THE INTERNATIONAL ASSOCIATION FOR PATTERN RECOGNITION

www.iapr.org

Volume 39, Number 1 NewsletterFebruary 2017 Special Issue with highlights of the 23rd International Conference on Pattern Recognition IssueIN THIS Letter from the President

Letter from the President Simone Marinai CALLS for PAPERS It is an honor for me to have IAPR event, in particular the ICPR Calls from IAPR Committees been elected President of the (International Conference on Pattern IAPR. In the next two years, I Recognition), a unique opportunity. Getting to Know... will have the privilege of working It is non-trivial to recall that the “I” Petia Radeva, IAPR Fellow with an exceptional Executive in IAPR stands for International. Committee: Massimo Tistarelli (1st IAPR...The Next Generation: Within the IAPR, scholars from a Vice President), Edwin Hancock Sungmin Eum wide range of countries freely take (2nd Vice President), Alexandra ICPR Highlights: on different responsibilities: holding Branzan Albu (Secretary), Apostolos From the General Chair offices, serving on committees, Antonacopoulos (Treasurer), and Traditional Dance of Mexico organizing events that take place in the Past President Ingela Nyström. Coffee Break/Lunch for Women nearly every area of the world. This is a great team, and I am sure Plenary Talk Abstracts that we will do our best to serve our With the admission of two new Invited Talk Abstracts association. members (Pakistan and Vietnam) Workshops/Tutorials/Contests there are now 49 member societies 2016 IAPR Fellows The IAPR has been my scientific represented on the Governing ICPR 2016 Awards family since I first participated in an Board. And, in 2018, the IAPR will 2016 Governing Board Meeting IAPR conference in 1995. The broad celebrate the 40th anniversary range of research topics covered by Bulletin Board of its incorporation and the first the IAPR makes participation in any Meeting and Education Planner meeting of its Governing Board, The views expressed in this newsletter represent the personal views of the authors and not necessarily those of their host institutions or of the IAPR. CALLS for PAPERS For the most up-to-date information on IAPR-supported conferences, workshops and summer schools, please visit the IAPR web site: www.iapr.org/conferences/

SSB 2017 DGCI 2017 14th IAPR/IEEE International Summer School on 20th International Conference on Discrete Geometry Biometrics for Computer Imagery Alghero, Italy Vienna, Austria Deadline: Feb. 15, 2017 Deadline: Mar. 1, 2017 Dates: Jun. 12-16, 2017 Dates: Sep. 19-21 2017 CVIP 2017 2nd International Conference on ICDAR 2017 Computer Vision and Image Processing 14th International Conference on & Workshop on Multimedia Document Analysis and Recognition Noida, India Kyoto, Japan Deadline: Mar. 1, 2017 Deadline: Mar. 15, 2017 Dates: Sep. 9-12, 2017 Dates: Nov. 10-15, 2017 PReMI 2017 ICPRS 2017 7th International Conference on Pattern Recognition 8th International Conference on and Machine Intelligence Pattern Recognition Systems Kolkata, India Madrid, Spain Deadline: Apr. 3, 2017 Deadline: Apr. 9, 2017 Dates: Dec. 5-8, 2017 Dates: Jul. 12-13, 2017

CIARP 2017 IJCB 2017 22nd Iberoamerican Congress on 3rd International Joint Conference on Biometrics Pattern Recognition Denver, Colorado, USA Valparaíso, Chile Deadline: Apr. 15, 2017 Deadline: May 21, 2017 Dates: Oct. 1-4, 2017 Dates: Nov. 7-10, 2017

PSIVT 2017 8th Pacific Rim Symposium on Image Analysis and Video Technology Wuhan, China Deadline: TBA Dates: Nov. 20-24, 2017

IAPR Then and Now... Excerpt from K-S Fu's CHAIRMAN'S CORNER LETTER IAPR Newsletter Vol. 1 No. 1 Aug. 1978 During the Second (1974) IJCPR in Copenhagen, the Standing Committee approved the proposal of establishing a permanent international professional organization for pattern recognition. After two year's preparation, the IAPR was formally established at the Third (1976) IJCPR at Coronado, California, and the Standing Committee approved the Constitution of IAPR and elected the Officers of the Executive Committee. Three committees, Publication, Conference, and Membership were appointed for each. [...] A special committee drafted the IAPR Bylaws which were then approved by the Herb Freeman, King-Sun Fu, Toshiyuki Sakai Standing Committee early in 1977. [...] IAPR was formally Incorporated under the laws of and Theo Pavlidis at the 4th IJCPR in Kyoto the State of New York in January 1978. (1978)

IAPR Newsletter, Vol. 39 No. 1, Feb. 2017 Page 2 Return to Page 1 which was held on November 7, ICPR Liaison Committee along aim at getting new (young) people 1978, during the Fourth IJCPR with all other concerned Standing involved. (International Joint Conference Committees as well as the leaders In the first meeting of the 2016- on Pattern Recognition) in of the Technical Committees. 2018 ExCo in Cancún, we planned Kyoto, Japan. From the list of 13 We aim to have some concrete several tasks for the Standing “founding” members, the Federal proposals on the table for a Committees. There is no need to Republic of Germany stands out, broader discussion during the next list all these tasks here, and I will a reminder of a period when walls ICPR in Beijing. just mention some relevant topics. divided Europe. In a time where Another area where I would like to new physical and virtual walls • The Conferences & Meetings make some progress is the support are being built around the world, Committee (C&M), in addition of the work of young researchers. I find it to be extremely important to the exceptional work It is not just rhetoric to say that that one scientific, learned society done to assess requests for young researchers are the future still considers an international sponsorship of conferences, of any learned society, and the sharing of responsibilities as a is expected to follow-up on IAPR is no exception. We should distinguishing feature engraved some open issues addressed continue to focus our resources in its DNA. And it is worth noting in the last term related to and efforts to support students to that it is a requirement for IAPR standardization of URLs and be involved in IAPR events, make sponsored and endorsed events to website archiving. Moreover, connections, and establish long- welcome participants from all IAPR the C&M will need to take care term professional relationships. Member Societies without any of some ICPR-related matters • Since ICPR 2002, the IAPR restriction due to political, racial, or like satellite workshops and has provided travel stipends to religious differences. reorganization of the ICPR young investigators to attend guidelines. Since the early years, the IAPR ICPR. • The Education Committee will has been closely intertwined • In the last two years, we have continue to handle the research with the ICPR (initially IJCPR). channeled IAPR support for scholarship applications that This is the flagship event of the Summer Schools towards we expect will grow in number association and the place where the significant reduction of in the coming years. the Governing Board meets attendance costs for students. • The most important task for biennially and where many • We also began, thanks to the ICPR Liaison Committee researchers have the opportunity the work of the Education will be to oversee an analysis to discuss their work with Committee, a new IAPR of potential changes in the colleagues (who often are also Research Scholarship program organization of ICPR. The friends). Within an increasingly that helps young researchers to aim is to improve the quality competitive research world there is visit research institutes in other (actual and perceived) of a need for researchers to publish countries. the conference, continuing the results of their work in widely All of these activities must to consider inclusion and recognized venues. Considering continue—and be strengthened— participation as highly that the broad range of topics in the coming years in a important features for ICPR. covered in PR is at the same time sustainable way, taking into • The Publications and a strength and a weakness, we account the IAPR’s financial Publicity Committee has been need to do our best to ensure that constraints. asked to propose possible ICPR is among those venues and updates in the IAPR Website to maximize the quality of IAPR The 15 active Technical and to continue the interactions publications in general. Committees are the scientific with publishers of IAPR backbone of the IAPR. Once In this light, I see of paramount journals. again, it is worth recalling what importance the need to re-think the the “I” in IAPR stands for. A Before I close this letter, I want to ICPR format. It is too ambitious to TC should be international in thank two people who work behind expect to come to any real change spirit and spread over several the scenes to make the IAPR in fewer than two years; however, IAPR member societies both for engine work. For many colleagues I would like to seed one concrete its research topics and for its Linda O’Gorman is just the kind discussion on the future of ICPR, leadership. Leadership rotation is person offering advice about our involving first and foremost the also important, with a continuous society at the IAPR booth during

IAPR Newsletter, Vol. 39 No. 1, Feb. 2017 Page 3 Return to Page 1 IAPR Then and Now Argentina joined in 2010 38 A look at IAPR Australia joined in 1991 104 membership at the Austria joined in 1985 182 beginning and today Belarus joined in 1993 67 Belgium joined in 1978 left in 2004

Column 1 shows the list of countries that Brazil joined in 2003 34 have, over the years, formed national Bulgaria joined in 1990 47 pattern recognition societies and joined the Canada joined in 1978 95 IAPR. Chile joined in 2007 25 Column 2 shows which societies formed the China joined in 1985 199 association in 1976 when the association Cuba joined in 2000 69 was formally established at the 2nd Czech Republic joined in 1993 62 International Joint Conference on Pattern Denmark 48 Recognition in Copenhagen. The IAPR was Finland 375 officially incorporated in 1978 with 13 France 351 member societies. Germany (GDR pre-1990) joined in 1985 390 Column 3 shows the IAPR's membership in Federal Republic of Germany one society from Germany after 1990 2016. With 49 member societies, the IAPR has over 12,000 members from around the Greece joined in 2002 37 world. Hong Kong joined in 1993 97 Hungary joined in 1985 74 ICPRs. Rather, in her role as India joined in 1987 158 IAPR Secretariat, Linda helps in Iran joined in 2010 168 an infinite number of matters from Ireland joined in 1999 61 handling tax forms and related Israel joined in 1985 27 bureaucratic issues to reminding Italy 366 me to write these notes. The Japan 559 second person I want to thank Korea (South) joined in 1992 202 is Ed Sobczak, who, in his role Macau joined in 2012 37 as IAPR Webmaster, efficiently Malaysia joined in 2014 25 and effectively manages the Mexico left in 1990; reformed/joined in 2004 91 IAPR website and performs other Morocco joined in 2003 left in 2009 related services that are probably Netherlands 250 unknown to many IAPR members. New Zealand joined in 1999 120 Lastly, I would like to thank the Norway joined in 1984 99 Governing Board once again for Pakistan joined in 2016 51 my election. I will serve, together Poland joined in 1993 92 with my ExCo colleagues, the Portugal joined in 1987 67 IAPR community with my best Russian Federation joined in 1989 1087 effort, and I look forward to seeing Singapore joined in 2004 55 you around the world in the next Slovenia joined in 1993 46 two years or at least in Beijing for South Africa joined in 1994 91 ICPR 2018! Spain joined in 1985 173 Simone Marinai Sweden 205 IAPR President 2016-2018 Switzerland joined in 1986 46 [email protected] Taiwan joined in 1994 367 Tunisia joined in 2014 75 Turkey joined in 1999 25 Ukraine joined in 1993 74 United Kingdom 542 USA 4608 www.iapr.org Uruguay joined in 2012 34 Vietnam joined in 2016 34

IAPR Newsletter, Vol. 39 No. 1, Feb. 2017 Page 4 Return to Page 1 Calls from IAPR Committees

From the IAPR Education Committee: From the IAPR Call for Applications Executive Committee IAPR Research Scholarships (ExCo): http://www.iapr.org/docs/IAPR-EC-RS-Call-2016.pdf Description: IAPR Research Scholarships, awarded by the IAPR Call for Proposals for through its Education Committee (IAPR-EC), seek to make possible mobility across institutions and international boundaries for Early Summer Schools Career Researchers working in fields within the scope of the IAPR’s Deadline: February 28, 2017 interests. Through this program, the IAPR sees an opportunity to make a (for schools planned for significant contribution to the development of Early Career Researchers April - July 2017) as well as the wider Pattern Recognition community. Covered expenses, funding and duration: Summer schools are training activities where participants are exposed to the • The scholarship will cover round trip travel and basic living expenses latest trends and techniques in the • The visits will be no longer than 12 months in duration. particular pattern recognition field. They provide a unique opportunity to Requirements: engage students and junior researchers • The candidate must be a full-time researcher (PhD research student with senior scientists in a fruitful way . who has completed at least one year’s study at this level or someone To be eligible for a grant, the already employed as a full-time researcher who has been active organizers must work through at least in the field for fewer than eight years and is working at a level one of the IAPR’s technical committees equivalent to a post-doctoral researcher. as they develop and present the • The candidate must be member of an IAPR member society. proposal. • The covered travel and housing expenses cannot be funded by another scholarship. If there is a shortfall between the actual costs Of course, the term “Summer School” and the amount covered by the Scholarship, the candidate may seek is somewhat generic and traditional. complementary funding, usually from either the home or the host There is no requirement that a school institution. be offered during the summer. • The host institution must be different from the candidate’s home How to Submit: Proposals for IAPR institution and should be in a different country. funded summer schools should • The home and host institutions must give explicit approval by a be submitted to IAPR Treasurer signed letter. Apostolos Antonacopoulos by email • A successful applicant will be permitted to adopt the title “IAPR ([email protected]). International Scholar” for the period of the award. A PDF attachment containing all the Click here for the full Call for Applications. required information is appreciated. Contact information: For detailed guidelines on the proposal, see the ExCo Initiative on IAPR-EC Chair IAPR Secretariat Summer Schools. c/o Josep Lladós c/o Linda O’Gorman [email protected] [email protected]

IAPR Newsletter, Vol. 39 No. 1, Feb. 2017 Page 5 Return to Page 1 Getting to know...Petia Radeva, IAPR Fellow

Can Egocentric Vision and Lifelogging help Healthcare?

Petia Radeva, IAPR Fellow ICPR 2016, Cancún for contributions to computer vision, machine learning, egocentric vision, life-logging, and medical imaging

Petia Radeva completed her undergraduate study at the University of Sofia, Bulgaria, in 1989; her Masters in 1993 and PhD in 1996 at the Autonomous University of Barcelona. Currently, she is an Associate Professor at the University of Barcelona (UB) and a Senior Researcher in the Computer Vision Center, Spain. She is Head of the Consolidated Research Group on Computer Vision at UB and Head of the Medical Imaging Laboratory of the Computer Vision Center (www.cvc.uab.es). by Petia Radeva, Head, Consolidated Research Group on Computer Her research interests are on development Vision (University of Barcelona) and Medical Imaging Laboratory of the of learning-based approaches for Computer Vision Center computer vision and health applications: End-to-end technologies for large scale Pattern Recognition and Computer Vision (PR&CV) are challenging image analysis; Using autobiographic fields by definition. Claiming to be able to see and understand an episodes captured by a lifelogging camera image, or being able to extract knowledge and learn from a sample to enhance memory of patients with mild of data as human beings do is a brave if not overoptimistic goal. It cognitive impairment; Monitorizing by has been overoptimistic from one of its first known Computer Vision a wearable camera to help people eat works, when one of the co-founders of the Artificial Intelligence Lab at better and thus assure healthier lifestyle; MIT in 1966 Marvin Minsky asked one of his Measuring active aging of older people. undergraduate students at MIT to implement a computational system to be able to see SCIENTIFIC MERITS: h-index of 36 and understand an image. During the last (Google Academic), 379 international 50 years, millions of scientific hours have publications, more than 90 impact factor been spent trying to solve this task. A lot of journals (Microsoft Academic Research), people had to step on the shoulders of giants achieving 5505 citations. in order to construct the PR&CV community, AWARDS: Petia Radeva was awarded where thirst for knowledge; collaboration; IAPR Fellow (2016), CIARP award Aurora cooperation; sharing of knowledge, articles, Pons Porrata (2016) and Academia data, algorithms, and work; and, not least, the ICREA (2014), award of the Catalonian very human wish to build something helpful to government to the best researchers in any others were a must to construct the PR&CV scientific field. tower similar to Catalonia’s human ones. Photo: https://en.wikipedia.org/ wiki/Castell IAPR Newsletter, Vol. 39 No. 1, Feb. 2017 Page 6 Return to Page 1 In fact, belonging to this community per month and 900,000 images per for 25 years, now I cannot year. Today, there is no problem IAPR Fellow Q & A determine which was the stronger to acquire and store such an motivation for me (maybe both): amount of information, but the real With their years of experience, thirst for knowledge or wish to problem is who will look at 900,000 IAPR Fellows have a lot to offer construct something to be helpful images?! How can the semantics to younger researchers just getting to society. Probably because of of this large amount of information started in their careers. With that this, I spent my Masters and PhD be extracted to make it useful? in mind, I asked Dr. Radeva some thesis on Computer Vision and additional questions. Medical Imaging. People using wearable cameras need a PR&CV community ~ Arjan Kuijper, EiC Going further with exploring effort to process and analyze how PR&CV algorithms can this amount of data, extract EiC: In some of your recent help healthcare, I thought about patterns of behavior and lifestyle, paper titles you raise questions - wearable cameras. Wearable extract information about social do you think that choosing these cameras are very recent and at interactions, activities, events, and titles attracts more attention (& the same time an exponentially habits in order to get real profit citations)? growing market tendency. from it. And the PR&CV community According to Tractica, the wearable is getting aware of this. Just look PR: I strongly believe that each camera market is still in the at the rapidly growing number of article should answer a question. early stage of development and articles on first-person camera In fact, this is one of my first has experienced exponential (egocentric) vision over the last 5 recommendations to PhD students, growth in the last few years, years as well as the workshops when writing a paper: what is the expanding from 5.6 million in organized at highly prestigious question each article addresses, 2014 to a predicted 30.6 million conferences like CVPR, ECCV, even what is the question each units annually in only three years ICCV, BMVC, ACM MM, the paragraph answers. Today, with from now (2020). Today, most number of special issues in the exponentially growing number of the wearable cameras (e.g. journals, and public datasets and of papers, messages in the articles GoPro) are used for leisure and code. should be clear and concise. So sport. My hypothesis is that such being able to define the main devices can have huge impact What are the problems and question is very helpful. technologies to be applied for healthcare, too. In particular, if EiC: Is deep learning a hype? last year 21 million users bought when processing large-scale Fitbit Fitness Trackers to help them images coming from wearable PR: Taking into account the number stay in healthy physical condition, cameras? Looking back 15 of works on deep learning, articles wearable cameras can give very to 20 years, we can see that published, successful systems important information about the our algorithms were not very (NVidia web page is reporting nutritional habits of a person, for ambitious: they had to recognize really interesting success stories example. This is really important, a small number of categories: of deep learning), companies are since recent nutritional studies AR Database(126), Yale investing big hope and efforts in show that what people eat could Face(15), Caltech(6), 101(101), this technology. Still, we should be as important as how they eat. VOC2005(10), TU Graz(4), not forget the pre-deep learning As an example, in a TedxMileHigh VOC2006(10), Caltech256(256), time. And trying to understand talk, Lauren Constantini explains MIT-CSAIL(125), VOC2007(21), the deep learning techniques and the case of a person significantly Cifar(10), Cifar-100(100). On the their relation to the techniques we reducing his weight by up to 45 other hand, taking into account have been developing decades of kg just by changing the social and the amount of data (several years before is really interesting physical conditions and context thousands of images), the goal and helpful to understand this of his meals (lighting and social was really challenging (and technology. On the other hand, context). perhaps, overoptimistic). Recently, we need for the technology to go there were two milestones that through the hype stage in order to The problem is that wearable made a tipping point in the arrive at the plateau of productivity, cameras used for lifelogging every image analysis: the advances in when the technology really will find day can take up to 2,500 images deep learning and, in particular, its place in the science and society. that means around 75,000 images convolutional neural networks

IAPR Newsletter, Vol. 39 No. 1, Feb. 2017 Page 7 Return to Page 1 (CNN) and the appearance of the analyze the thousands of images enjoyable cognitive exercises to ImageNet database with 1,000 about people wearing the camera keep our brain fit and detect the categories and 15 million images. and their environment; that will be most early age-related mental CNN could go so far due to the able to give them advice about problems. fast improvement of hardware how and what they eat in order to We are almost there to complete allowing processing millions of keep them healthy; that will be able the undergraduate project of images with models of millions of to diagnose early when people are Minsky’s student: being able to parameters. Being helped by the getting too much stress or even understand an image. But will we GPUs processing, today we are are entering into the first stages of be able to monitor and understand able to experiment with different a depression, just by observing the ourselves? Today, more than ever, architectures of CNN having up evolution of their social interaction. our PR&CV community has all to 150 layers. Today, we have And when we get older, the the necessary tools and a wide the hardware and software tools algorithms should be able to follow spectrum of open problems to to experiment with complex our active aging, keeping track of work hard on in order to advance architectures in a "Lego®" fashion. how we communicate, how we do scientifically and achieve a real And although we can “design our basic activities like nutrition technological breakthrough for neural networks that are up to or resting, or our instrumental science, society and humanity. 10 times cheaper and with up to activities like being able to shop, First and very promising results 15 times less parameters that at to transport, to communicate by are here. Let’s maintain the strong the same time outperform their phone, and even create a very collaboration between research predecessors from one year ago, personalized serious games with and industry, and keep working the next challenge in the near our proper images by creating hard in this direction! future is to have them working on our mobile devices without extra hardware support or prohibitive memory overhead” (Christian Szegedy, Senior Research Scientist at Google). But the real challenge in Deep learning is how to take advantage of the millions of unlabeled and unstructured data in order to improve deep learning algorithms and create systems working in real situations. In the same way, as 10 years ago face detection and recognition were among the most challenging and difficult problems, today face detection is applied in every camera or image-based social media, we expect that in the next five years PR&CV algorithms will be able to understand social interactions, understand and summarize video clips, recognize video and human activities, do text processing or extract medical experts knowledge from medical images being basic for screening, diagnosis, intervention and follow- up. Photo: Diego Sánchez/Victor Alvarez And in this direction, we hope that Ingela NystrÖm presenting the IAPR Fellow Award to Petia Radeva deep learning will also help with lifelogging data to extract and

IAPR Newsletter, Vol. 39 No. 1, Feb. 2017 Page 8 Return to Page 1 IAPR...The Next Generation

In this series of Feature Articles, the IAPR Newsletter asks young researchers to respond to three questions:

• Briefly: How did you get involved in pattern recognition? • In more detail: What technical work have you done and what is/are your current research interest(s)? • How can the IAPR help young researchers?

~Arjan Kuijper, Editor -in-Chief

Sungmin Eum Sungmin Eum is a PhD candidate by Sungmin Eum, Department of Electrical and getting ready to defend his work from Computing Engineering, University of Maryland, the Department of Electrical and College Park, USA Computer Engineering, University of Maryland, College Park, U.S.A. Briefly: How did you get involved in pattern He is advised by Dr. David recognition? Doermann at the Institute for It all started when I got a chance to work on the Advanced Computer Studies “The detection of Diphthongs for knowledge- (UMIACS). He was the based speech recognition” for my undergraduate winner of the Zamperoni independent study. This came to me as a Best Student Paper Award significant point of intersection where my area of at the International study could meet the beauty of nature. Since I Conference on was a person who greatly relished the wonders Pattern Recognition of human bodies and nature, the process of (ICPR) 2016. His finding hidden patterns in natural signals certainly research interest provided me with pure delights and stronger includes various motivations that I could not find in any other topics which bridge areas. This was when I decided to get involved computer vision, in a research that closely dealt with patterns from machine learning human beings and nature. To satisfy my thirst and (particularly in to equip myself with substantial skills, I decided deep neural nets), to join the masters program without a second computational thought. photography and image processing I started off as a research assistant in the which are motivated Computer Vision Laboratory in Yonsei University, by human-inspired ideas. Korea. I participated in the project, “Abnormal User Detection Using Face Occlusion Verification for Automatic Teller Machines (ATM)”. As this Editor's note: work was partially funded by a corporate vendor Sungmin Eum was the winner of the Piero Zamperoni Best who wanted to commercialize the product at Student Paper Award at ICPR 2016. The conference is the subject the end of the day, I had to deal with double of this ICPR 2016 Special Issue of the IAPR Newsletter. challenges concurrently. I had to come up with ~ Arjan Kuijper, Editor-in-Chief an algorithm with state-of-the-art performance, but with low computational complexity at the

IAPR Newsletter, Vol. 39 No. 1, Feb. 2017 Page 9 Return to Page 1 same time. While working on this frames. We specifically target the tries to integrate the learned project, I remember getting my first cases where the highlights are information from architecturally English-written paper accepted to saturated and the original contents different tasks within a single present at the CVPR Workshops. are completely obscured. The unified framework to boost up It certainly was an overwhelming method is based on an interesting the primary task performance. opportunity for me, not only to feel observation that when two Moreover, the investigation of the joy of sharing my work with images are captured at different human actions is also one of the others, but also getting to know viewpoints, the displacement topics I find intriguing. Specifically, that there were lots of researchers of the target content is different I am currently seeking to break who are willing to share how they from that of the highlight regions down the human actions into fine- “recognize the patterns” in their (Motion parallax). This work was grained action atoms which can be unique domains. meaningful for its first use of generally used and shared across ‘highlight correspondences’ for different actions. In more detail: What technical better localization of the highlights work have you done and what How can the IAPR help young in multiple images. is/are your current research researchers? interest(s)? My very recently published work Whenever I get an opportunity to was about selecting frames of Currently, my PhD research could attend a conference, I always get a planar object in a video or a be described largely into two more than what I expect to get. set of images by analyzing their streams. The first line of research Not only does it bring a chance to “frontalness”. In my paper, I mainly focuses on ‘camera-based learn new techniques and ideas defined "frontalness" as a measure image and video analytics’ that from the presentations, but also it of how close the surface normal mixes a bit of computer vision provides time and space to freely of an object aligns with the optical techniques and some flavor of interact with others where I can get axis of a camera. The unique and image processing. My first problem valuable opinions and feedback novel aspect of our method is that was to tackle the challenge of about my work and thoughts. unlike previous planar object pose capturing high-quality images What’s important is to be able to estimation methods, our method of large sources in a single leave a relatively small research does not require a frontal reference frame with limited field-of-view, society where people often revolve image. Our approach was particularly for document images. around similar ideas and to meet motivated by an observation that a For this problem, I proposed a with people who have been true frontal image can be used to novel graphcut-based image struggling to overcome entirely reproduce other non-frontal images mosaicking method which seeks to different challenges. Listening by perspective projection, while the overcome the known limitations of to their ideas and learning from non-frontal images have limited the previous approaches. Basically, their thought processes, which ability to do so. the method guides a cut in the are often different in nature, helps overlapping portion of the two As the second line of research, me to step back and observe my images so that the cut line would I have recently been working problems with a fresh perspective. make as little artifact as possible. on image/video classification A conversation with a person who Our method does not require any and object detection problems majors in bioinformatics even led prior knowledge of the content using deep learning approaches, to one of my recent publications. of the given document images, specifically using the deep It doesn't necessarily have be a making it more widely applicable convolutional neural networks conference. What matters is a and robust. One of the interesting (CNN). One of my recent works forum for young researchers to parts is that we incorporated a was to find out how semantic share their ideas with the others sharpness measure which induces keywords (e.g., police, fire, who have different perspectives. cut generation in a way that smoke, helmet, and car) and I believe this is where the IAPR results in the mosaic including the visual features could come could come into play. The IAPR sharpest pixels. together to better classify events could help in setting up such that may look similar but contain The second problem I addressed occasions from time to time in semantically different situations. was the issue of removing the different regions. A steady and I am also delving into the same highlight regions caused by the well-established program would event recognition problem with a light sources reflecting off glossy certainly be of great asset to all the rather different approach, which surfaces, for instance, picture young “isolated” researchers.

IAPR Newsletter, Vol. 39 No. 1, Feb. 2017 Page 10 Return to Page 1 Cancún, Mexico, December 4-8, 2016 Highlights

Comments from the General Chair Comments from the Traditional Dances of Mexico ICPR 2016 General Chair Coffee Break/Lunch for Women @ ICPR Photo : Rangachar Kasturi Since the creation of the IAPR in the 1970s, this was the first time that Plenary Lecture Abstracts: an ICPR was held in Latin America, and we were very proud to welcome K. S. Fu Prize the ICPR 2016 participants to Cancún, a remarkable, original, and Robert Haralick, beautiful city on the Yucatán Peninsula, known not only for its beaches J. K. Aggarwal Prize and the ancient Maya culture but also for its traditional downtown area. Fei-Fei Li, We are sure that the attendees enjoyed the conference and the cultural and richness of the city. Maria Petrou Prize Michal Irani The novelty of this edition of the conference was that we brought technological advances to Latin America for the benefit of research Invited Talk Abstracts centers, universities, and society as a whole, hoping that it will be a • Josien Pluim fundamental stimulus for progress in science and education for the • Wolfgang Förstner near future. Also, we took measures to encourage women to participate • Ricardo Baeza-Yates in all activities promoted by the IAPR as–for the first time–we offered • Marc Pollefeys a coffee break and lunch for female scientists (47 female researchers participated). Workshop Reports A total of 1,234 manuscripts were submitted to the conference. Following the review process, 199 papers were accepted for oral presentations Tutorial Reports and 478 were accepted as posters. In addition, 33 contests and four invited papers were included in the proceedings. The technical program Contest Reports included 40 oral sessions distributed in five parallel tracks, four poster sessions, and seven plenary sessions including the K-S Fu, the J. K. 2016 IAPR Fellows Aggarwal and, for the first time, the Maria Petrou lectures. 933 delegates from 50 countries attended ICPR 2016. ICPR 2016 Awards: • BIRPA and Zamperoni The organization of such a large conference was the result of the • Intel Best Scientific Papers contribution of many people. Our gratitude goes to members of the • IBM Best Student Papers Program Committee and Track Chairs who dedicated their time to the • ICPR 2016 Distinguished review process and to the preparation of the program. We also thank the Performance Award reviewers who have evaluated the papers and provided the authors with • IAPR Certificate of valuable input on their research work. Appreciation Finally we acknowledge the work of conference committee members • Elsevier Awards who strongly contributed to make this event successful and, I hope, unforgettable. 2016 Meeting of the IAPR Governing Board Eduardo Bayro-Corrochano General Chair, ICPR 2016 CINVESTAV,Campus Guadalajara, Mexico

IAPR Newsletter, Vol. 39 No. 1, Feb. 2017 Page 11 Return to Page 1 ICPR 2016 Highlights

Traditional Dances of Mexico at ICPR 2016

Editor's note: At the banquet, ICPR 2016 delegates and guests were treated to a spectacular demonstration of traditional dances of Mexico. ~ Arjan Kuijper, Editor-in-Chief

"Folk dances are dances that are musicians in studded charro developed by people that reflect outfits and wide-brimmed hats the life of the people of a certain • Jarabe Tapatío (Mexican Hat country or region." Dance), which was a courting ~Wikipedia article on Folk dance dance Mexican dance has been • Son Jarocho, from Veracruz, influenced by the country's Mexico, with a blend of African, cultural, political and religious Spanish and native rhythms history. Some of the themes that • La Danza del Venado (Deer come through are related to the Dance), a ritualistic dance from Mexican Revolution, the Spanish the Yaqui region of Mexico with Colonial Period, Catholicism, and dancers in the roles of deer Nature. and hunter The photos on this page show • Tlacolorerosis (agricultural several types of music and dance), from the Guerrero dancing, including: region, with roots in the Aztec • Mariachi, the cultural and religion spiritual tradition marked by

Photos: Diego Sánchez/Victor Alvarez

IAPR Newsletter, Vol. 39 No. 1, Feb. 2017 Page 12 Return to Highlights Return to Page 1 ICPR 2016 Highlights Coffee Break for Women and Lunch for Women @ ICPR Tuesday, December 6, 2016 Recognizing the steadily growing number of female researchers in fields of interest to the IAPR and hoping to foster new connections among these researchers, the IAPR Executive Committee and the ICPR2016 organizers sponsored two events for the first time: a Coffee Break for Women at ICPR and a Lunch for Women at ICPR.

Both of these events followed the first Maria Petrou Lecture—given by Michal Irani—that honored a "female scientist/engineer who has made substantial contributions to the field of Pattern Recognition (or a closely related field), and whose past contributions, current research activity and future potential may be regarded as a model to both aspiring and established researchers".

Photo: Susan Vincil, www.vistaimagery.com Lunch for Women at Mocambo Restaurant

~

Approximately 30 women attended the luncheon

Photos: Diego Sánchez/Victor Alvarez IAPR Newsletter, Vol. 39 No. 1, Feb. 2017 Page 13 Return to Highlights Return to Page 1 ICPR 2016 Highlights Winner of the 2016 K. S. Fu Prize Robert Haralick City University of New York, USA

For contributions in image analysis including remote sensing, texture analysis, mathematical morphology, consistent labeling, and system performance evaluation.

Dependency has multiple forms, each of which in its own way restricts possibilities. If xNx1 is a discretely valued random N-tuple governed by a probability P, the uncertainty about a random value of x is the entropy associated with P. If the uncertainty is less than the maximal possible uncertainty for x, that decrease in uncertainty induces a stochastic dependency among the components of x and/or makes certain values of certain components more or less probable. The most visual case of stochastic dependency occurs in the image domain. Any patch of an image that shows a texture is a region having a stochastic dependency among the pixel values of the patch. The gray level co-occurrence matrix captures a second order dependency among the values of neighboring pixels. Functionals of the co-occurrence matrix can be used as features in distinguishing one texture from another. We show how the commonly used can be improved by putting the co-occurrence matrix into the setting of graphical models and how it is possible to produce a texture transform image. Each pixel in a textural transform image gives the joint occurrence probability of the neighborhood values centered around that pixel. The size of the neighborhood can be arbitrary. Another class of dependencies is driven by relationships, for example spatial relationships. A consistent labeling approach permits recognizing objects that are always in a legal spatial relationship. We review how a consistent labeling problem can be posed and then define a relation join-decomposition generalization. Solving the relation join-decomposition is equivalent to learning relationships in data. It is interesting that learning relationships necessarily involves learning relationships of no-influence. No-influence obeys the same semi-graphoid properties that conditional independence obeys. Structure specifies the organization of data. One level of organization is the partitioning of data into clusters, each of whose points are similar to one another. The classic K-means algorithm uses a structure of zero- dimensional manifolds. Similarity is specified by the distance of a point to the cluster center. Clustering can be done where the ideal of each cluster is a k-dimensional manifold where k can vary from cluster to cluster. Here similarity can be specified by the distance of the point to the manifold. eW give an algorithm for doing such clustering and show how it can be used prior to doing regression to get more meaningful regression relationships. IAPR Then and Now...20 Years Ago, IAPR Newsletter Vol. 18 No. 4, Oct. 1996 Excerpts from Dr. Maria Petrou's From the Editor's Desk column following the 13th ICPR in Vienna, Austria, at which Prof. Robert Haralick was elected IAPR President

Dear Everybody, And when you thought it was safe to forget all about Vienna and 13th ICPR, you got this issue and it brought it all back! This newsletter is dominated by the activities in Vienna and it comes to you hot from the press! In fact it is so hot, that some cookies did not manage to get into the oven before the door shut! So, the report on the technical content of ICPR as well as the New President's address will be included in the January issue of the newsletter. I can of course report that from August 22 this year our Big Boss is Professor Haralick, elected as president by the Governing Board in Vienna. Apart from this change, our Executive Committee has a new 1st Vice President, Professor Gelsema, a new 2nd Vice President, Professor Kidode, while the Treasurer remains the same, Dr. Bigun, and the secretary Dr. Sanniti di Baja is also the same, and of course Professor Kittler remains on the Committee as the Past President. IAPR Newsletter, Vol. 39 No. 1, Feb. 2017 Page 14 Return to Highlights Return to Page 1 ICPR 2016 Highlights Winner of the 2016 J. K. Aggarwal Prize Fei-Fei Li Stanford University, USA

For seminal contributions to object and event classification using learning methodologies based on statistical and neuroscience principles.

It takes nature and evolution more than five hundred million years to develop a powerful visual system in hu- mans. The journey for AI and computer vision is about fifty years. In this talk, I will briefly discuss the key ideas and the cutting edge advances in the quest for visual intelligences in computers. I will particularly focus on the latest work developed in my lab for both image and video understanding, powered by big data and the deep learning (a.k.a. neural network) architecture.

Editor's note: ICPR 2016 saw the inauguration of the IAPR's Maria Petrou Prize, which is given "to a living female scientist/engineer who has made substantial contributions to the field of Pattern Recognition (or a closely related field), and whose past contributions, current research activity and future potential may be regarded as a model to both aspiring and established researchers". http://www.iapr.org/fellowsandawards/awards_petrou.php. Please also see photos from the Coffee Break for Women and Lunch for Women @ ICPR, both events were held for the first time at ICPR 2016. ~ Arjan Kuijper, Editor-in-Chief Winner of the 2016 Maria Petrou Prize Michal Irani Weizmann Institute for Science, Israel

For pioneering contributions to space-time video analysis, motion estimation, and image analysis by composition and as a role model for early career researchers striving for excellence and rigour in the fields of computer vision and pattern recognition. Title: “Blind” Visual Inference In this talk I will show how “blind” visual inference can be performed by exploiting the internal redundancy inside a single visual datum (whether an image or a video). The strong recurrence of patches inside a single image/video provides a powerful data-specific prior for solving complex tasks in a “blind” manner. The term “blind” here is used with a double meaning: (i) Blind in the sense that we can make sophisticated inferences about things we have never seen before, in a totally unsupervised way, with no prior examples or training data; and (ii) Blind in the sense that we can solve complex Inverse-Problems, even when the forward degradation model is unknown. I will show the power of this approach through a variety of example problems (as time permits), including: 1. “Blind Optics” — recover optical properties of the unknown sensor, or optical properties of the unknown environment. This in turn gives rise to Blind-Deblurring, Blind Super-Resolution, and Blind-Dehazing. 2. Segmentation of unconstrained videos and images. 3. Detection of complex objects and actions (with no prior examples or training).

IAPR Newsletter, Vol. 39 No. 1, Feb. 2017 Page 15 Return to Highlights Return to Page 1 ICPR 2016 Highlights

Invited talk in Biomedical Image Analysis and Applications

Josien Pluim, Groningen University, The Netherlands Validation of image registration: the truth is hard to make

Medical image registration has been a topic of research for many decades by now. In those years, enormous progress has been achieved in terms of the quality of the solutions, the complexity of problems that can be tackled and the speed of computation. There is one aspect falling far behind with respect to progress: validation. Although attempts to validate image registration methods have been made, their number and success do not match that of method development. Finding ways to evaluate method performance is a very tough problem. The presentation will focus on ways in which registration is standardly validated, on recent developments in validation and on dilemmas encountered. Although some aspects treated are specific to image registration or medical applications, many of the material covered is generic and holds for other areas of image analysis.

Invited talk in Image, Speech, Signal and Video Processing

Wolfgang Förstner, Institute of Geodesy and Geoinformation, Bonn, Germany A Future for Learning Semantic Models of Man-Made Environments

Deriving semantic 3D models of man-made environments hitherto has not reached the desired maturity which makes human interaction obsolete. Man-made environments play a central role in navigation, city planning, building management systems, disaster management or augmented reality. They are characterised by rich geometric and semantic structures. These cause conceptual problems when learning generic models or when developing automatic acquisition systems. These conceptual problems are caused by the interplay between (1) the elementary data structures of the 3D objects and the observed images or 3D point clouds, (2) the topology of the rich spatial structures, (3) the partonomy and taxonomy of complex objects, and (4) the uncertainty and qualitativeness of the used notions in the envisaged application domains. I will sketch some research issues in conceptual modelling playing a key role for a further evolution of acquiring man-made environments from visual data.

IAPR Newsletter, Vol. 39 No. 1, Feb. 2017 Page 16 Return to Highlights Return to Page 1 ICPR 2016 Highlights

Invited talk in Pattern Recognition and Machine Learning

Ricardo Baeza-Yates, Universitat Pompeu Fabra, Barcelona, Spain Visual Congruent Ads for Image Search

The quality of user experience online is affected by the relevance and placement of advertisements. We propose a new system for selecting and displaying visual advertisements in image search result sets. Our method compares the visual similarity of candidate ads to the image search results and selects the most visually similar ad to be displayed. The method further selects an appropriate location in the displayed image grid to minimize the perceptual visual differences between the ad and its neighbors. We conduct an experiment with about 900 users and find that our proposed method provides significant improvement in the users’ overall satisfaction with the image search experience, without diminishing the users’ ability to see the ad or recall the advertised brand. This is joint work with Yannis Kalantidis, Ayman Faharat, Lyndon Kennedy and David A. Shamma.

Invited talk in Computer Vision

Marc Pollefeys, ETH Zurich and Microsoft, Zurich, Switzerland Semantic 3D reconstruction Obtaining 3D geometric information from images is one of the big challenges of computer vision. It is critical for applications such as robotics, autonomous vehicle navigation and augmented reality. I will first briefly talk about some our recent work on vision-based autonomous micro-aerial vehicles, driverless cars and 3D on mobile devices. While purely geometric models of the world can be sufficient for some applications, there are also many applications that need additional semantic information. Next, I will focus on 3D reconstruction approaches which combine geometric and appearance cues to obtain semantic 3D reconstructions. Specifically, the approach I will present is formulated as a multi-label volumetric segmentation, i.e. each voxel gets assigned a label corresponding to one of the semantic classes considered, including free-space. We propose a formulation representing raw geometric and appearance data as unary or high-order (pixel- ray) energy terms on voxels, with class-pair-specific learned anisotropic smoothness terms to regularize the results. We will see how by solving both reconstruction and segmentation/ recognition jointly the quality of the results for both subtasks can be improved and we can make significant progress towards 3D scene understanding.

IAPR Newsletter, Vol. 39 No. 1, Feb. 2017 Page 17 Return to Highlights Return to Page 1 ICPR 2016 Highlights

ICPR 2016 Workshops

Contents for this section and IWPRHA3 links to other sections in the 3rd International Workshop on ICPR2016 Highlights: Pattern Recognition for Comments from the General Chair Healthcare Analytics Traditional Dances of Mexico https://sites.google.com/site/iwprha3/ Coffee Break/Lunch for An ICPR 2016 Workshop, endorsed by the IAPR Women @ ICPR December 4, 2016, Cancún, Mexico Plenary Lecture Abstracts: Invited Talk Abstracts Workshop Organizers: Workshop Reports: Faisal Farooq, IBM, USA IWPRHA3 (PR for Healthcare Jianying Hu, IBM, USA Analytics) Stein Olav Skrøvseth, University Hospital of North Norway MPRSS (Multimodal PR for Rogerio Abreu De Paula, IBM, Brazil Social Signal Processing in Human Interaction) By Faisal Farooq PRRS (PR in Remote Sensing) Introduction MANPU (CoMics ANalysis, IWPRHA3 brought together pattern recognition and healthcare Processing and Undertanding) researchers interested in healthcare analytics and applications of UHA3DS (Understanding pattern recognition in this field. The workshop program consisted of Human Activities presentations by invited speakers from pattern recognition and from through 3D Sensors) healthcare and by authors of papers submitted to the workshop. In addition, there was an interactive panel discussion to identify important FFER (Face and Facial problems, applications and synergies between the pattern recognition Expression Recognition) and healthcare analytics disciplines. The intended audience of the DLPR (Deep Learning for PR) workshop included pattern recognition researchers interested in solving healthcare analytics problems as well as the healthcare community in CVAUI (CV for Analysis of general including payers, providers and researchers. Underwater Imagery) Workshop Content RRPR (Reproducible Research in PR) Keynote Talks Tutorial Reports The workshop featured two keynote talks delivered by internationally renowned researchers in the field. Contest Reports 2016 IAPR Fellows The first keynote talk was delivered by Prof. Robert Jenssen of the University of Tromsø–The Arctic University of Norway. Prof. ICPR 2016 Awards Jenssen directs the Machine Learning group at the university and in 2016 Meeting of the IAPR recent years has heavily focused on developing machine learning Governing Board and pattern recognition algorithms for healthcare applications. Prof.

IAPR Newsletter, Vol. 39 No. 1, Feb. 2017 Page 18 Return to Highlights Return to Page 1 ICPR 2016 Highlights Jenssen discussed Data-Driven for the workshop and each Decision-Tree Prediction of Colorectal Surgical submission was reviewed by at Panel Complications using Machine least three experts in the field. Learning and Pattern Recognition. The final decision to select six The workshop concluded with an In his keynote talk he described papers for presentation was interactive panel. The panelists efforts to leverage recent machine based on the cumulative scores included the two keynote speakers learning and pattern recognition each submission received. Each Profs. Jenssen and Prabhakaran methodology for predicting presentation included 20 minutes as well as one of the workshop postoperative complications of presentation and five minutes of organizers Dr. Jianying Hu. The in colorectal surgery. Such Q&A. The presentations included a panel was moderated by Dr. Faisal complications include anastomosis very good mix of machine learning, Farooq. The panelists answered a leakage, a potentially lethal pattern recognition and image wide variety of questions including condition, wherein early detection processing technologies on but not limited to: is very important. 1. Learning similarities between 1. Why is the time right for Data The second keynote talk was irregularly sampled short Science to be mainstream in delivered by Prof. B. Prabhakaran multivariate time series from healthcare? of the University of Texas at EHRs Dallas. Prof. Prabhakaran’s 2. What are some unique 2. Evaluation of face-to-face talk title was “All Living Things challenges healthcare poses? communication skills for Move” – Kinematics in Human people with dementia using a 3. Where are the gaps in the Healthcare and Performance. Prof. head-mounted system problems that are being Prabhakaran discussed the recent addressed by the community developments in body sensor 3. Extracting Important Factors and what can be done to technology in capturing human of Prescribing Medicine bridge those gaps? body and body part motions, and by Data Analysis of Health the techniques for analyzing the Checkup and Tele-Medical 4. Where are the big areas of captured data. He also described Intervention Program in opportunity in healthcare that his experiences with applications Developing Countries Data Science can address? in fields such as Physical Medicine 4. Texture Features for The panelists as well as the & Rehabilitation, Medical Physics audience shared their experiences & Radiation Oncology, and Speech Classification of Vascular Ultrasound including challenges and success Language Pathology. stories which served as cues 5. Unsupervised Fully Automated Contributed Talks for various questions and a Cartilage Segmentation from productive discussion. Many young The workshop also included Knee MRI researchers including students, six paper presentations that scientists, faculty in this field found 6. Nucleus-based Cell were selected for presentation these extremely useful as they Classification in Cervical after a peer review process. We are marching on a journey in this PAP Smear Images using a received a total of 10 submissions discipline.

Robert Jenssen, University of TromsØ, B. Prabhakaran, University of Texas (Dallas) Norway Title: All Living Things Move - Kinematics in Human Health-care Title: Data-Driven Prediction of Colorectal and Performance Surgical Complications using Machine Learning Abstract excerpt: "[...] In this talk, we discuss the recent developments in body and Pattern Recognition sensor technology in capturing human body and body part motions, and the Abstract excerpt: "[...] The talk will describe efforts to techniques for analyzing the captured data. We also describe our experiences leverage recent machine learning and pattern recognition with applications in fields such as Physical Medicine & Rehabilitation, Medical methodology for predicting postoperative complications in Physics & Radiation Oncology, and Speech Language Pathology. Then, we colorectal surgery. Such complications include anastomosis consider the advances in mixed/augmented/virtual reality and explore possible leakage, a potentially lethal condition, wherein early applications incorporating them along with the sensors used for capturing detection is very important." human kinematics.

IAPR Newsletter, Vol. 39 No. 1, Feb. 2017 Page 19 Return to Highlights Return to Page 1 ICPR 2016 Highlights

MPRSS 2016 4th International Workshop on Multimodal pattern recognition of social signals in human computer interaction https://neuro.informatik.uni-ulm.de/MPRSS2016/ An ICPR 2016 Workshop, December 4, 2016, Cancún, Mexico

Organization: Friedhelm Schwenker, Ulm University, Germany Stefan Scherer, University of Southern California, USA by Friedhelm Schwenker Cognitive Technical Systems at theoretical and applied aspects. Ulm University and the Technical The fourth edition of the MPRSS Committee on Pattern Recognition Selected and revised papers will workshop was organized by Dr. in Human-Machine Interaction be published in Summer 2017 in Friedhelm Schwenker (Institute (TC9) of the International the Springer LNCS/LNAI series as of neural Information Processing, Association for Pattern Recognition postconference proceedings. Ulm University, Germany) and (IAPR). Prof. Stefan Scherer (Institute of In addition to the regular program Creative Technologies, University For the MPRSS 2016 workshop an invited tutorial on Multimodal of Southern California, USA). 18 papers were submitted from Recognition of Mental States: The MPRSS 2016 workshop was which 10 were selected for oral Emotions, Dispositions, Pain was sponsored by Ulm University, the presentation at the workshop. given by Harald C. Traue, Steffen University of Southern California, Papers presented original research Walter, Holger Hofmann, Sascha the Transregional Collaborative in machine learning and pattern Meudt, Friedhelm Schwenker all at Research Center SFB/TRR recognition in human computer Ulm University. 62 Companion-Technology for interaction focusing on both

ICPR 2016 participants at the 4th International Workshop on Multimodal Pattern recognition of social signals in human computer interaction (HCI)

Photos: Diego Sánchez/Victor Alvarez

IAPR Newsletter, Vol. 39 No. 1, Feb. 2017 Page 20 Return to Highlights Return to Page 1 ICPR 2016 Highlights PRRS-9-2016 9th Workshop on Pattern Recognition in Remote Sensing http://iapr-tc7.de/prrs/PRRS2016.htm An ICPR 2016 Workshop, endorsed by the IAPR, December 4, 2016, Cancún, Mexico

Workshop Chairs: Eckart Michaelsen, Germany Jie Shan, USA by the Workshop Chairs from satellite or aerial images. Often these data are taken in The IAPR's Technical Committee 7 different spectral bands than on Remote Sensing and Mapping the usual RGB, and in particular organizes workshops on pattern more recent data usually have recognition applied to remotely more than just three colors. In sensed data. This has quite some fact so-called hyper-spectral data tradition by now, and it is regularly have become very popular in done in close temporal and remote sensing, featuring about geographical neighborhood of the two hundred bands. E.g. the key- ICPRs. note of this PRRS (held by D. In 2016 the PRRS was again held Tuia, Univ. of Zurich) was mostly in the form of a one-day workshop motivated by such hyperspectral Uwe Stilla (chair ISPRS ICWG II/III), Eckart Michaelsen (chair IAPR-TC7, and within the frame of the ICPR, this applications. Basically the spectral chair PRRS-9-2016), Jie Shan (co-chair time in Cancún, Mexico, on the bands suitable to remote sensing are given by the atmospheric IAPR-TC7, and co-chair PRRS-9-2016), Sunday before the start of the main and Devis Tuia (IEEE-GRSS and keynote conference. transmission. So, next to the visual and the infra-red domain, also the speaker of PRRS-9-2016) Remotely sensed data, such as RADAR domain is an important satellite and aerial imagery, set source, in particular satellite- Xplore. The speakers were from one of the classical application based synthetic aperture RADAR diverse countries such as the USA, topics of pattern recognition. But (known as SAR). As additional China, Germany, Brazil, Russia, the workshop is also intended to source there is LIDAR (air-based France etc. Since the workshop draw experts and researchers from or mounted on vehicles). This is was held in conjunction with the the photogrammetry and remote where considerable overlap with ICPR, the audience was very sensing communities to the ICPR the robotics community exists. international as well, and there and to make them aware of the Next to pixel-wise classification were more than just the speakers, IAPR–fostering inter-disciplinary object recognition is also a topic, key-note speaker, and chairs cooperation. This PRRS–like concentrating on things like listening. In fact there were about most of its predecessors–was co- buildings, vehicles, roads, or rivers. forty people present, and technical sponsored by IAPR-TC7, ISPRS discussions were very vivid. ICWG II/III, and IEEE-GRSS, and The 14 technical talks held during the workshop can be assigned Whoever is interested in these most of the participants actually topics and in the next PRRS (to are more active in the remote to the topics outlined above. These works were selected from be held in China in 2018) should sensing communities than in the contact Eckart Michaelsen (TC7 IAPR. 17 submitted manuscripts on the base of single blind reviews by the Chair, eckart.michaelsen@iosb. The classical topic of pattern proven and diligent TC-7 program fraunhofer.de) or Jie Shan (TC7 recognition in remote sensing is committee. The corresponding Co-chair, [email protected]), and pixel-wise land-use classification proceedings will appear in IEEE- join IAPR-TC7.

IAPR Newsletter, Vol. 39 No. 1, Feb. 2017 Page 21 Return to Highlights Return to Page 1 ICPR 2016 Highlights

An ICPR 2016 Workshop, endorsed by the IAPR http://manpu2016.imlab.jp/

General Co-chairs: Jean-Marc Ogier, University of La Rochelle, France Kiyoharu Aizawa, The University of Tokyo, Japan Koichi Kise, Osaka Prefecture University, Japan

Program Co-chairs: Jean-Christophe Burie, University of La Rochelle, France Proceedings are availble in the ACM Toshihiko Yamasaki, The University of Tokyo, Japan Digital Library. Motoi Iwata, Osaka Prefecture University, Japan by the General Co-chairs and the The 12 accepted papers and one The steering committee for Program Co-chairs invited paper were published by the MANPU decided to submit ACM in ACM Digital Library. The the workshop proposal of the MANPU 2016 received URL is as follows: http://dl.acm. 2nd International Workshop on contributions from seven countries. org/citation.cfm?id=3011549 coMics ANalysis, Processing The number of submitted papers and Understanding to ICDAR and accepted papers were 17 Prof. Akihiko Shirai from 2017 (International Conference and 12, respectively. The review Kanagawa Institute of Technology on Document Analysis and process was carried out by the (Japan) was invited to give keynote Recognition). Program Committee that consisted talk titled “Manga generator, a of 18 outstanding researchers, future of interactive manga media: The next edition of the workshop all of whom are specialists of invited talk paper.” The paper of will be held in Kyoto, Japan in comics analysis, processing and his keynote talk is in public in ACM November 2017. understanding. Digital Library at the following URL: http://dl.acm.org/citation.cfm?id=30 MANPU 2016 consists of one 15156&CFID=719897196&CFTOK invited talk, two oral sessions EN=79955713 (total 6 presentations), one poster session (total 6 presentations) and one panel discussion in a single track. The presentations widely covered the topics related to comics researches, for example, the dataset of comics, comics analysis, comics image processing, comics understanding, comics visualization, and so on. The number of attendee was around 40. IAPR Newsletter, Vol. 39 No. 1, Feb. 2017 Page 22 Return to Highlights Return to Page 1 ICPR 2016 Highlights

http://www-rech.telecom-lille.fr/uha3ds2016/

Workshop Chairs: Mohamed Daoudi, Télécom Lille, France Francisco Flórez-Revuelta, University of Alicante, Spain Pietro Pala, University of Firenze, Italy Hazem Wannous, University of Lille 1, France by the Organizing Committee The review process was carried and body language. out by the Program Committee, The objective of UHA3DS 2016 composed of experts in the The invited speaker, Prof. Rita was to bring together researchers workshop topics who provided Cucchiara from Università di from computer vision and machine three independent reviews for Modena e Reggio Emilia Italy, learning communities, working each submitted paper. Only 9 gave a keynote speech entitled together in a natural synergy papers were accepted as oral "Human behavior understanding & and having an interest in using presentations. 3D data". recent computing technologies to understand humans, but also The Workshop program consisted Proceedings of UHA3DS 2016 will support them. About 25-30 people of two sessions covering topics be published in the Springer LNCS attended the workshop, resulting in such as 3D pose estimation, Series. valuable scientific exchanges. human activity analysis, hand gesture analysis, body expression

Rita Cucchiara (Ms.Electr. Eng. ’89, Phd in Comp. Eng. ’92) is full professor at UNIMORE, Italy. She is Director of the Interdip. Research Center in ICT “Softech-ICT”, Director of Master in “Visual Computing and Multimedia Technology” and coordinates the Research Lab Imagelab. She is Fellow and Member of the Governing Board dell’Intern. Ass. of Pattern Recognition IAPR since 2012 and Member of the Advisory Board of the Computer Vision Foundation since 2016. The Research activities cover Computer Vision, Pattern Recognition and Machine Learning, Multimedia, for video analysis by surveillance cameras, human behaviour understanding, people detection and tracking and social interaction recognition. She coordinated several research projects: among them: Italy- France Galileo project with INRIA Paris (2008-09); BESAFE , funded by NATO “Science for Peace”, with Hebrew Univ. Israel (2008-12); THIS, European CHIPO/JLS project MIUR

Italian Technology Clusters in smart cities and communities (2014-17) She is involved Alvarez /Victor Sánchez Diego Photo: in activities of scientific organization; among them: Area Chair at CVPR 2014, 2016, Program Chair at ICCV 2017, Tutorial chair at ECCV2016,ACMMM2016,ICPR2016. MPRSS 2016 Invited Speaker, Prof. Rita Rita Cucchiara is author of more than 270 papers in international journals and proceeding Cucchiara, Università di Modena e Reggio of international conferences with referees; She has currently H-INDEX=37, citation 8100, Emilia Italy i-10index 134.

IAPR Newsletter, Vol. 39 No. 1, Feb. 2017 Page 23 Return to Highlights Return to Page 1 ICPR 2016 Highlights

FFER 2016 2nd International Workshop on Face and Facial Expression Recognition from Real World Videos http://www.vap.aau.dk/ffer16/ An ICPR 2016 Workshop, December 4, 2016, Cancún, Mexico Organizers: Kamal Nasrollahi, Aalborg University, Denmark Gang Hua, Stevens Institute of Technology, USA Thomas B. Moeslund, Aalborg University, Denmark Qiang Ji, Rensselaer Polytechnic Institute, USA by Kamal Nasrollahi expression recognition that are The proceedings of the workshop being carried out at Professor will be published by Springer. The workshop received eight Pietikäinen’s center for Machine submission, out of which seven Vision Research. There were about were accepted for the oral 50 people attending the workshop. presentation. Besides these seven presentations, the technical The presented papers included program included a keynote topics like, Classification of speech which was given by Subtle Facial Expressions, Face Professor Matti Pietikäinen from Recognition, Facial Expression the University of Oulu in Finland. Recognition, Age Estimation, Face Dr.Sc.Tech. Professor Matti Pietikäinen The speech highlighted the Verification, Head Pose Estimation, gave a keynote on Towards Reading research works on face and facial and Super-resolution. Emotions from Face Videos.

DLPR 2016 1st International Workshop on Deep Learning for Pattern Recognition http://dlpr2016.csp.escience.cn/dct/page/1 An ICPR 2016 Workshop, December 4, 2016, Cancún, Mexico Workshop Chairs: Xiang Bai (Huazhong University Science and Technology), Zhaoxiang Zhang (Institute of Automation, Chinese Academy of Sciences), Shiguang Shan (Institute of Computing Technology, Chinese Academy of Sciences), Chunhua Shen University of Adelaide), Yi Fang (New York University), Jingdong Wang (Microsoft Research Asia), Yangqing Jia (Facebook), Shuicheng Yan (National University of Singapore) 35 submissions to DLPR were the workshop and took part in the In addition, parts of accepted received, and 20 of them were related discussions. papers were invited to a "Pattern accepted after peer reviews Recognition Letters" Special Issue The pdf files of all the accepted performed with easychair, including on: Deep Learning for Pattern papers have been published on the six oral papers and 14 poster Recognition (DLPR). workshop website: http://dlpr2016. papers. Around 150 researchers csp.escience.cn/dct/page/1. or graduate students attended

IAPR Newsletter, Vol. 39 No. 1, Feb. 2017 Page 24 Return to Highlights Return to Page 1 ICPR 2016 Highlights

http://cvaui2016.oceannetworks.ca/

An ICPR 2016 Workshop, December 4, 2016, Cancún, Mexico

Workshop Chairs: Alexandra Branzan Albu (University of Victoria, BC, Canada) Maia Hoeberechts (Ocean Networks Canada, BC, Canada) Photo: Diego Sánchez/Victor Alvarez by the Workshop Chairs automated analysis. Due to the talks were given by prominent properties of the environment itself, researchers from ocean science, Monitoring marine and freshwater the analysis of underwater imagery industry and computer science ecosystems is of critical imposes unique challenges which perspectives. Dr. Henry Ruhl, importance in developing a better need to be tackled by the computer National Oceanography Centre, understanding of their complexity, vision community in collaboration Southampton, UK, spoke on including the effects of climate with biologists and ocean “Advances in Computer Vision and change and other anthropogenic scientists. Pattern Recognition for Research influences. The collection of in Biological Oceanography.” Dr. This workshop was well attended underwater video and imagery, Anthony Hoogs, Kitware, Clifton (30 participants) and provided a whether from stationary or moving Park, NY, USA, talked about the forum for researchers to share platforms, provides a non-invasive “Video and Imagery Analytics for and discuss new methods and means of observing submarine the Marine Environment (VIAME): applications for underwater ecosystems in situ, including the an Open Source Framework for image analysis. We received 14 behavior of organisms. Underwater Image Processing.” submissions, out of which 10 were Finally, Dr. Yogesh (Yogi) Girdhar, Oceanographic data acquisition accepted based on a thorough Woods Hole Oceanographic has been greatly facilitated by the peer review process. Many of the Institution (WHOI), Woods Hole, establishment of cabled ocean submitted papers were of excellent MA, USA, entertained workshop observatories, whose co-located quality, so the high acceptance participants with his talk entitled, sensors support interdisciplinary rate reflects a self-selection “Real-time Unsupervised Scene studies and real-time observations. process performed by the authors Understanding for Building Curious Scheduled recordings of of the submissions. We thank the Underwater Exploration Robots.” underwater video data and static members of Program Committee images are gathered with Internet- for lending their time and expertise We hope that all workshop connected fixed and PTZ cameras, to ensure the high quality of the attendees will be inspired in which observe a variety of accepted workshop contributions. their research by participating biological processes. These cabled in CVAUI 2016, and that this This year’s technical program ocean observatories, such those workshop will foster many fruitful covered a variety of topics, operated by Ocean Networks conversations and open new areas including applications to fisheries Canada (www.oceannetworks.ca), for collaborative interdisciplinary research, species classification, offer a 24/7 presence, resulting in research in underwater image and scene reconstruction, unprecedented volumes of visual analysis. data and a “big data” problem for among others. Three keynote IAPR Newsletter, Vol. 39 No. 1, Feb. 2017 Page 25 Return to Highlights Return to Page 1 ICPR 2016 Highlights

Workshop Chairs: Miguel Colom (CMLA, ENS Cachan, France) Bertrand Kerautret (LORIA, Université de Lorraine, France) Pascal Monasse (LIGM, École des Ponts ParisTech, France) Jean-Michel Morel (CMLA, ENS Cachan, France) by the Workshop Chairs poster. We thank the scientific of the authors who obtained the committee, allowing high quality Reproducible label is archived A new event was proposed with works including papers, algorithms and publicly available in the the first edition of the Workshop on and source code reviews. A total of GitHub account of the organizing Reproducible Research in Pattern 12 accepted papers (including the committee, https://github.com/ Recognition. The aim of this two invited papers) are going to be RLPR. workshop was to give an overview published as post-proceedings by Actually, seven papers were of Reproducible Research (RR) for Springer in the LNCS series. authors, with a special focus on submitted and two are already Pattern Recognition algorithms. The different topics of workshop accepted and the others are under were equitably represented peer-review. More submissions The call for papers was organized with six presentations for the might be accepted. More details into two main tracks: RR RR Framework track and five are available at the RRPR website Frameworks and RR Results. presentations for the RR Result Finally a selection of papers will be The first track was dedicated to track. Two invited presentations invited to an RRPR IPOL special the general topic of Reproducible were given during the workshop. issue (www.ipol.im). Research in Computer The first one was a presentation Science with papers describing of the Image Processing On Line experiences, frameworks and journal (IPOL) presented by Pascal platforms. The second track Monasse in a common work focused on the description with Miguel Colom. The second of previous works in terms of invited talk was given by Daniel Reproducible Research. The latter Lopresti and Bart Lamiroy with a track contained ICPR companion presentation of the DAE platform papers describing their quality of in the context of Reproducible Reproducible Research. Research. The number of This workshop received 16 attendees were around 30. submissions with eight papers For its first edition, submitted to Track 1 of RR the RRPR committee Framework, six papers to Track introduced the 2 of RR Results, and two papers "Reproducible associated to the invited talks. Label in Pattern After a reviewing process including Recognition" in order to highlight mostly three reviewers per paper, the reproducible aspects of the six papers were accepted as RRPR and ICPR works. The work oral presentations and four as IAPR Newsletter, Vol. 39 No. 1, Feb. 2017 Page 26 Return to Highlights Return to Page 1 ICPR 2016 Highlights

ICPR 2016 Tutorials

Adversarial Pattern Recognition Contents for this section and links to other sections in the ICPR2016 Highlights: An ICPR 2016 Tutorial, December 4, 2016, Cancún, Mexico Comments from the General Chair Lecturer: Traditional Dances of Mexico Fabio Roli, IAPR Fellow, IEEE Fellow (University of Cagliari, Italy) Coffee Break/Lunch for Women @ ICPR "If you know the enemy and know security and privacy tasks, pattern Plenary Lecture Abstracts: yourself, recognition techniques will be Invited Talk Abstracts you need not fear the result of a soon targeted by specific attacks, hundred battles" crafted by skilled attackers. In Workshop Reports ~ Sun Tzu, The art of war, 500 BC particular, two main threats against Tutorial Reports: learning algorithms have been Adversarial Pattern Recognition Learning-based pattern classifiers identified, among a larger number are currently used in several of potential attack scenarios, Handling Blur applications, like biometric respectively referred to as evasion Multimodal human behavior recognition, spam filtering, and poisoning attacks. This kind analysis in the wild malware detection, and intrusion of problem has been named detection in computer networks, adversarial pattern recognition, Data-driven Pattern which are different from traditional and is the subject of an emerging Recognition pattern recognition tasks. The research field in the machine Image-Based Measurement difference lies in the fact that in learning community. these applications an intelligent, Similarity searching adaptive adversary can actively The purposes of this tutorial were: Contest Reports manipulate patterns with the aim of (a) to introduce the fundamentals 2016 IAPR Fellows making a classifier ineffective. of adversarial machine learning to the ICPR community; (b) to ICPR 2016 Awards Traditional machine learning illustrate the design cycle of a techniques do not take into 2016 Meeting of the IAPR learning-based pattern recognition Governing Board account the adversarial nature system for adversarial tasks; (c) of classification problems like to present the new techniques recognition tasks like biometric the ones mentioned above. One that have been recently proposed recognition and spam filtering. of the consequences is that the to assess performance of pattern About 40, coming from different performance of standard pattern classifiers under attack, evaluate classifiers can significantly countries, attended, and there classifiers’ vulnerabilities, and was a lively discussion, which degrade when they are used implement defense strategies that in adversarial tasks. Pattern witnesses the interest for this topic make learning algorithms and in our community. classifierscan be significantly pattern classifiers vulnerable to well-crafted, more robust against sophisticated attacks exploiting attacks; (d) to show knowledge of the learning and some applications of classification algorithms. adversarial machine Being increasingly adopted for learning to pattern

IAPR Newsletter, Vol. 39 No. 1, Feb. 2017 Page 27 Return to Highlights Return to Page 1 ICPR 2016 Highlights

Handling Blur

An ICPR 2016 Tutorial, December 4, 2016, Cancún, Mexico

Lecturers: Jan Flusser, Filip Šroubek, and Barbara Zitová all three lecturers are from the Institute of Information Theory and Automation, Czech Academy of Sciences, Czech Republic

Blur is an unwanted phenomenon, quality. However, image restoration In the tutorial, we focused on both which is common in digital methods are often ill-posed, ill- approaches. We started with blur images. It results in smoothing conditioned, and time consuming. modeling and analyzed potential high-frequency details, which sources of blur in real images. On the contrary, the blur-invariant makes the image analysis difficult. In the first part of the tutorial we approach (yellow in Figure), Heavy blur may degrade the reviewed traditional as well as proposed originally in 1995, works image to such an extent, that modern deconvolution techniques, directly with the blurred data neither automatic analysis nor including blind deconvolution, without any preprocessing. Blurred visual interpretation of the content space variant deconvolution, and image is described by features, are possible. If we did not have multichannel deconvolution. In which are invariant with respect proper tools for processing and the second part we discussed to convolution with some group analyzing blurred images, many invariants to image blurring. of kernels. Image analysis is then unique images would be lost. Two The tutorial was completed with performed in the feature space. major approaches to handling numerous demonstrations and This approach is suitable for object blurred images exist. They are practical examples. recognition, template matching, more complementary rather and other tasks where we want to The attendance of around than concurrent; each of them is recognize/localize objects rather 10 people was relatively low. appropriate for different tasks and than to restore the complete However, the lively discussion employs different mathematical image. The mathematics behind it during the coffee break and at the methods and algorithms. is based on projection operators end of the tutorial, indicated the Image restoration (blue in Figure) and moment invariants. deep interest of the attendees. is one of the oldest areas of image processing. It appeared as early as the 1960's and 1970's in the work of the pioneers A. Rosenfeld, H. Andrews, B. Hunt, and others. In the last ten years, this area has received new impulses and has undergone a quick development. We have witnessed the appearance of multichannel techniques, blind techniques, and superresolution enhancement resolved by means of variational calculus in very high-dimensional spaces. A common point of all these methods is that they remove the blur from the input image and produce an image of a high visual Figure. Two major approaches to handling blurred images exist: image restoration (blue) and blur-invariant (yellow)

IAPR Newsletter, Vol. 39 No. 1, Feb. 2017 Page 28 Return to Highlights Return to Page 1 ICPR 2016 Highlights

Multimodal human behavior analysis in the wild: recent advances and open problems

An ICPR 2016 Tutorial, December 4, 2016, Cancún, Mexico

Lecturers: Xavier Alameda-Pineda, INRIA Grenoble Rhône-Alpes Nicu Sebe, University of Trento

During the IAPR International During the second part of the and surveillance. Special emphasis Conference on Pattern course, a more in-depth analysis was given to recent approaches Recognition (ICPR) of 2016, in of a few of the methodologies was combining signal processing Cancún, we had the honor to give given. In particular, we described and machine learning to robustly a tutorial course on multimodal how low-rank matrix completion extract crucial low and middle level human behavior analysis in the approaches can be used to information. wild. We made a special effort to address different tasks with the The tutorial was instructed by Prof. describe recent advances and robustness required to operate in Nicu Sebe (University of Trento) open problems in the field, by scenarios with noisy and missing and by Dr. Xavier Alameda-Pineda showing practical examples and data, specific to applications in (INRIA Grenoble Rhône-Alpes). discussing the methodological the wild. Moreover, we discussed choices for different applicative variational Bayesian approaches We greatly acknowledge the efforts scenarios. that are able to reduce the of Dr. Elisa Ricci (Fondazione complexity of EM algorithms in Bruno Kessler and University of The course lasted four hours with probabilistic modelling with latent Perugia) in jointly preparing the a short break in the middle. The variables while keeping a good slides that she could not present first two hours were devoted to overall performance of the system. due to family constraints. We show a vast amount of examples missed you! for which off-the-shelf methods We designed the tutorial for would not work and different researchers from the pattern All the material is available online modifications needed to be done. recognition community interested at the URL: http://mhug.disi.unitn.it/ The applications that motivated in computer vision and audio tutorial-icpr-2016/. these advances are truly diverse, processing for human behavior for instance: eye tracking, extreme analysis applications. All along We believe the course tutorial was head-pose estimation, crowd the course we focused on the a success and we are willing to analysis, separation of moving automated analysis of human prepare another tutorial in related sound sources, detection of social behavior in unstructured scenarios topics in the upcoming ICPR in attention attractors or heart-rate and its many potential applications Beijing! estimation from unconstrained in health care, conflict and people face videos. management, sociology, marketing

www.icpr2018.org

IAPR Newsletter, Vol. 39 No. 1, Feb. 2017 Page 29 Return to Highlights Return to Page 1 ICPR 2016 Highlights Data-driven Pattern Recognition: Philosophical, Historical, and Technical Issues

An ICPR 2016 Tutorial, December 4, 2016, Cancún, Mexico

Lecturers: Marcello Pelillo, IAPR Fellow, IEEE Fellow (Italy) Fabio Roli, IAPR Fellow, IEEE Fellow (Italy) “How does it happen that a properly by a variety of initiatives and The Human Condition (1933) endowed natural scientist comes to concern studies such as a recent special René Magritte himself with epistemology? Is there no more issue of Pattern Recognition can’t have it run into your furniture valuable work in his specialty? I hear many Letters (October 2015). too many times. You don’t want it of my colleagues saying, and I sense it from to put your cat in the dishwasher many more, that they feel this way. I cannot The growing availability of even once […]. If you’re talking share this sentiment.” inexpensive data for some real- about a robot in a real-world ~ Albert Einstein (1916) world tasks, combined with the environment, you need for it to increasing power of computers, Paraphrasing a well-known learn things quickly from small brought naturally to the current remark by Carl von Clausewitz, amounts of data.” If the above data-driven paradigm, which can it can safely be said that “pattern remark is correct, and certainly is, be seen as a modern incarnation recognition is the continuation of then the business model based on of empiricist philosophy. The epistemology by other means.” “data bulimia” is not so profitable practical success of this approach In fact, fundamental questions as many claim, and it should be has raised the issue of “the pertaining to categorization, questioned. unreasonable effectiveness of abstraction, generalization, data” and led Google researchers The goal of this tutorial was to induction, etc., have been on to maintain that “memorization is critically address all these issues the agenda of mainstream a good policy if you have a lot of and provide pattern recognition philosophy, under different names training data.”1 If memorization is researchers, in particular the and guises, since its inception. a good policy, this seems to imply youngers who did not live the With the advent of modern digital that all the work done in the earlier early days of research, a picture computers and the availability of good days about generalization of the philosophical and historical enormous amounts of raw data, from small data sets and issues that are behind the current these questions have now taken a exploitation of a priori knowledge paradigm of Big Data + Deep computational flavor. is outdated. Pattern recognition Learning + GPUs, also discussing As it often happens with scientific can then be fully driven by (big) some technical issues that are research, in the early days of data, assuming that you can get relevant for a comprehension pattern recognition there used gigabytes of data very quickly, with that goes beyond the mainstream to be a genuine interest around no real cost. Unfortunately, there and trends. We felt that this could philosophical and conceptual are real-world tasks for which this be an opportunity for reflection, issues, but over time the interest is simply not possible. In a recent reassessment and eventually shifted almost entirely to technical Edge conversation2, Gary Marcus, some synthesis, with the aim of and algorithmic aspects, and a professor of psychology and a providing the field a self-portrait became driven mainly by machine-learning expert, said: “If of where it currently stands and practical applications. In recent you have a robot at home, you where it is going as a whole, years, however, there has been and hopefully suggesting new an increasing interest around 1A. Halevy, P. Norvig, and F. Pereira, The directions. the epistemological problems unreasonable effectiveness of data. IEEE Intel- ligent Systems, vol. 24, no. 2, pp. 8-12, 2009. The tutorial was attended of machine learning, from both by around 40 international 2https://www.edge.org/conversation/gary_mar- the computer scientist’s and the participants. The lively discussion philosopher’s camps, as witnessed cus-is-big-data-taking-us-closer-to-the-deeper- questions-in-artificial showed strong interest in this topic. IAPR Newsletter, Vol. 39 No. 1, Feb. 2017 Page 30 Return to Highlights Return to Page 1 ICPR 2016 Highlights

Image-based measurement

An ICPR 2016 Tutorial, December 4, 2016, Cancún, Mexico

Lecturer: Cris Luengo, Flagship Biosciences, Colorado, USA

In this tutorial, we discussed shown in most textbooks are case of linear structuring elements. several different topics related biased, the methods we discussed Although the tutorial did not attract to the task of obtaining are not. We analyzed (roughly) as many people as I had hoped, accurate (unbiased) estimates their precision. we did have a very engaged of morphological properties of The third part of the tutorial dealt audience, with lots of interesting real-world objects, as imaged with sampling and aliasing, and questions and discussions. I’m by a camera, microscope, MRI showed one method, called soft usually happy to have even one scanner, etc. This is a topic poorly clipping, of converting an intensity such student in my classroom! addressed in image processing image into a quasi-binarized image textbooks, even if the scientific I have written about unbiased in which there is much less aliasing literature has quite a bit to say measurements of binarized objects than a binary image. We then about it. on my blog: http://www.crisluengo. discussed a method to quantify net/index.php/archives/tag/ We first reviewed some basic area (volume), perimeter (surface measure. concepts from the field of area), bending energy and Euler stereology. In that field there is number from such quasi-binarized several hundred years’ experience images, yielding estimates several in deriving accurate estimates orders of magnitude more precise of morphological properties of than their binarized counterparts. objects, and many of the tools The last part of the tutorial and techniques used in that field showed how to obtain size translate very well to images. distributions without performing Some of these techniques are so any segmentation at all. We used important, for example how to deal the granulometry, an established with image edge effects, that they method from Mathematical should be taught in every image Morphology, to do so. We saw processing class. applications of the granulometry Next we discussed algorithms using isotropic structuring to quantify area (volume in 3D), elements (disks) as well as linear perimeter (surface area in 3D), structuring elements, and dis- Feret diameters and other features cussed the use of the path opening of binarized objects. Methods to improve performance for the Alvarez /Victor Sánchez Diego Photo:

IAPR Newsletter, Vol. 39 No. 1, Feb. 2017 Page 31 Return to Highlights Return to Page 1 ICPR 2016 Highlights

Similarity searching: Indexing, Nearest neighbor finding, Dimensionality reduction, and Embedding methods for applications in multimedia database

An ICPR 2016 Tutorial, December 4, 2016, Cancún, Mexico

Lecturer: Hanan Samet, University of Maryland, USA

as measured by the distance finding (see http://donar.umiacs. metric in the embedding space umd.edu/quadtree/index.html). The approximate the actual distance. second was the SAND Internet The search in the embedding Browser which enables browsing space uses conventional indexing a spatial relational database. methods which are often coupled The third was NewsStand which with dimensionality reduction. The enables reading news with a map key to the embedding methods is query interface which makes

Diego Sánchez /Victor Alvarez /Victor Sánchez Diego that the distance in the embedding use of spatial synonyms (see This half-day tutorial was space should be less than the newsstand.umiacs.umd.edu). true distance which makes the presented by Prof. Hanan Samet Professor Hanan Samet is embedding contracitve. This of the University of Maryland a Distinguished University means that if the distance of an at College Park. Attendance Professor at the University of object from the query object in started slowly and ramped up to Maryland. He has a Ph.D from the embedding space is greater approximately 50 attendees by the Stanford University. He is the conclusion of the tutorial. than the query distance, then the object is too far. Some commonly Prof. Samet started the tutorial known embedding methods are by pointing out that similarity multidimensional scaling, Lipschitz searching is usually achieved by embeddings, and FastMap. means of nearest neighbor finding. Existing methods for handling The tutorial was organized into similarity search in this setting fall five parts that covered the five into one of two classes. The first basic concepts outlined above: author of the text "Foundations of is based on mapping to a low- indexing low and high dimensional Multidimensional and Metric Data dimensional vector space which is spaces, distance-based indexing, Structures" published by Morgan- then indexed using representations dimensionality reduction, Kaufmann, an imprint of Elsevier, such as k-d trees, R-trees, embedding methods, and nearest San Francisco, 2006, "Design quadtrees, etc. The second directly neighbor searching. and Analysis of Spatial Data Structures" and "Applications of indexes the objects based on Prof. Samet concluded the tutorial Spatial Data Structures: Computer distances using representations with demos of existing systems Graphics, Image Processing and such as the vp-tree, M-tree, etc. that deploy some of the techniques GIS" published by Addison-Wesley, Mapping from a high-dimensional discussed in the tutorial. The first Reading, MA, 1990. space into a low-dimensional was the VASCO system which space is known as dimensionality enables the implementation of He received the 2011 ACM Paris reduction and is achieved using many spatial data structures Kanellakis Theory and Practice SVD, DFT, etc. At times, when we and seeing how they are built Award, which is one of the highest just have distance information, as well as used in algorithms for awards in Theoretical Computer the data objects are embedded operations such as windowing Science. in a vector space so that the and incremental nearest neighbor distances of the embedded objects

IAPR Newsletter, Vol. 39 No. 1, Feb. 2017 Page 32 Return to Highlights Return to Page 1 ICPR 2016 Highlights

Contents for this section and links to other sections in the ICPR 2016 Contests ICPR2016 Highlights: Comments from the General Chair Traditional Dances of Mexico Coffee Break/Lunch for Women @ ICPR Plenary Lecture Abstracts Invited Talk Abstracts December 4, 2016 Workshop Reports http://scenebackgroundmodeling.net/ Cancún, Mexico Tutorial Reports Contest Organizers: Contest Reports: Pierre-Marc Jodoin (Université de Sherbrooke, Canada) Scene Background Modeling Lucia Maddalena (National Research Council, Italy) Multimedia Challenges Beyond The aim of the SBMC 2016 contest Level Stationary Background Visual Analysis was to advance the development Generation Method Based on Subgraph Spotting in Graph of algorithms and methods for LaBGen". Representations of Comic scene background modeling The success of SBMC 2016 is Images through objective evaluation to be credited to the contribution on the SBMnet dataset, that PR Techniques for Indirect of many people, including the includes videos spanning different Immunofluorescence IA Program Committee Members: categories (basic, intermittent Thierry Bouwmans, Universitè Graph Distance motion, clutter, jitter, illumination La Rochelle (France), Maurizio changes, background motion, very Mobile Iris CHallenge Giordano, National Research long, and very short). Evaluation (MICHE-II) Council (Italy), Zhiming Luo, Detection/Recognition of Eight contest result and paper University of Sherbrooke Arabic Text in Videos submissions were accepted, (Canada), Alfredo Petrosino, authored by researchers of University of Naples Parthenope 2016 IAPR Fellows different countries, namely (Italy), Sébastien Piérard, ICPR 2016 Awards Belgium, China, France, Germany, University of Liège (Belgium), Yi Italy, Japan, Korea, Mexico, Qatar, Wang, University of Sherbrooke 2016 Meeting of the IAPR and Spain. The proposed methods (Canada). Moreover, the SBMnet Governing Board cover most of the methodologies dataset, the website and the commonly adopted for background utilities associated with this modeling and initialization, benchmarking facility would not based on temporal statistics, have materialized without the subsequences of stable Intensity, tireless efforts of Yi Wang and iterative model completion, missing Martin Cousineau (Université de data reconstruction, and neural Sherbrooke, Canada) and of the networks. Staff of the CVPRLab (University of Naples Parthenope, Italy). The Contest winners, Benjamin Laugraud, Sébastien Piérard and Marc Van Droogenbroeck, received a plaque for results achieved by From left to right, Lucia Maddalena, their method "LaBGen-P: A Pixel- Benjamin Laugraud (SBMC 2016 winner), Pierre-Marc Jodoin IAPR Newsletter, Vol. 39 No. 1, Feb. 2017 Page 33 Return to Highlights Return to Page 1 ICPR 2016 Highlights

Joint Contest on Multimedia Challenges Beyond Visual Analysis http://chalearnlap.cvc.uab.es/ An ICPR 2016 Contest, December 4, 2016, Cancún, Mexico Organizers: Hugo Jair Escalante (Mexico), Jun Wan (China), Sergio Escalera (Spain), Isabelle Guyon (USA), Michael Riegler (Norway), Xavier Baró (Spain), Henning Müller (Switzerland), Martha Larson (the Netherlands)

We organized an academic outstanding participation: almost The main goals of the challenge competition and workshop at 200 participants registered for were fulfilled: the state of the ICPR2016 that focused on four the contest, and 20 teams sent art was advanced considerably problems that require effective predictions in the final stage. in the four tracks, with novel processing of multimodal Results and winners of the contest solutions to the proposed problems information in order to be solved. were presented at ICPR. Overall (mostly relying on deep learning). Two tracks were devoted to the workshop had the participation However, further research is still gesture spotting and recognition of 11 speakers with published required. The data of the four from RGB-D video, two papers in the proceedings and tracks will be available to allow fundamental problems for human three outstanding keynote researchers to keep making computer interaction. Another track speakers: Alberto del Bimbo, Fabio progress in the four tracks. was devoted to a second round A. González and L. Enrique Sucar. Keep in touch with the ChaLearn of the first impressions challenge The contest was supported by LAP website for further events of which the goal was to develop three organizations with vast on looking at people. Please methods to recognize personality experience and prestige in the check the overview paper for traits from short video clips. For organization of academic contests, further details: H. J. Escalante et this second round we adopted a namely: Chalearn (http://www. al. ChaLearn Joint Contest on novel collaborative-competitive chalearn.org/), MediaEval (http:// Multimedia Challenges Beyond (i.e.,coopetition) setting. The www.multimediaeval.org/), and Visual Analysis: An overview. fourth track was dedicated to the ImageCLEF (http://www.imageclef. Proceedings of ICPR Workshops, problem of video recommendation org/). The contest was also 2016. for improving user experience supported by the IAPR TC-12 on under a certain watching situation. Multimedia and Visual Information The challenge was open for Systems (http://iapr-tc12.info/). about 45 days, and received

Prize award ceremony, team leaders for winners of each track: 1) Ali Salah (BU-NKU); 2) Yunan Li (FLiXT); 3) Xiujuan Chai (ICT- NHCI); 4) Mathias Lux (ITEC-AAU)

IAPR Newsletter, Vol. 39 No. 1, Feb. 2017 Page 34 Return to Highlights Return to Page 1 ICPR 2016 Highlights

SSGCI 2016 Competition on Subgraph Spotting in Graph representation of Comic Book Images

http://icpr2016-ssgci.univ-lr.fr An ICPR 2016 Contest, December 4, 2016, Cancún, Mexico

Organizers: Jean-Christophe Burie (University of La Rochelle, France) Pasquale Foggia (University of Salerno, Italy) Clément Guérin (University of La Rochelle, France) Thanh Nam Le (University of La Rochelle, France) Muhammad Muzzamil Luqman (University of La Rochelle, France) Jean-Marc Ogier (University of La Rochelle, France) Christophe Rigaud (University of La Rochelle, France) by Muhammad Muzzamil Luqman

Graphs are widely used for representing the structure, topology and attributes of underlying information in various application domains of pattern recognition. Information retrieval based on the structural (and topological) similarity between query and retrieval candidates can be best modeled by an attributed graph retrieval problem, which thus is a very important research problem especially for the application domains of structural pattern recognition, computer vision, image analysis, data mining and machine learning. This research problem becomes more challenging if the graphs contain attributes on their nodes and arcs. The research problem of searching a query graph in a database of graphs is termed as “subgraph spotting”. The proposed competition focused on the research problem of subgraph spotting in a database of attributed graphs. The goal of the SSGCI competition was to spot a query attributed graph in a database of attributed graphs. This means that for a given query attributed graph the goal is to retrieve every graph in the database that contains this query graph and to provide node correspondences between the query and each of the result graphs (as shown in the figure). This main challenge of the SSGCI competition Figure: Summarizing the ICPR 2016 Competition on Subgraph represented an open research problem in Spotting in Graph Representations of Comic Book Images (SSGCI

IAPR Newsletter, Vol. 39 No. 1, Feb. 2017 Page 35 Return to Highlights Return to Page 1 ICPR 2016 Highlights graph-based structural pattern pattern recognition: “Document IAPR Technical Committees recognition. The problems of Image Analysis, Recognition and TC10, TC11 and TC15 matching, indexing and retrieval Understanding”. We selected collaborated to organize the of graph-based representations the “comic book images” as the SSGCI 2016 competition. of underlying data are actively data source for extracting graph researched by various research representations, as they have lots teams around the globe. of relational information in their content and the various building The SSGCI competition was blocks of the comics exhibits organized as a joint collaboration challenging diversity in their shape of the various members of three and attributes while maintaining IAPR Technical Committees enough discriminatory information (TCs): IAPR TC-10 on GRAPHICS that permits their identification and RECOGNITION, IAPR TC- recognition. The use of comic book 11 on READING SYSTEMS images as the source for extracting and IAPR TC-15 on GRAPH the graph representations for the BASED REPRESENTATIONS. competition not only ensured that Researchers from L3i Laboratory there is enough variability in the of University of La Rochelle graph dataset, but it also ensured (France) and University of that there isn’t too much variability Salerno (Italy) participated in so that the graph matching the organization of the SSGCI becomes too difficult. The dataset competition. is freely available for the academic The dataset for the SSGCI research via the website of competition was obtained from a the SSGCI competition (http:// real problem domain of structural icpr2016-ssgci.univ-lr.fr).

Keywords cloud of the SSGCI competition

IAPR Newsletter, Vol. 39 No. 1, Feb. 2017 Page 36 Return to Highlights Return to Page 1 ICPR 2016 Highlights

Photo: Diego Sánchez /Victor Alvarez Brian C. Lovell (left) and Gennaro Percannella Contest Chairs: (right) on behalf of the Brian C. Lovell (U. Queensland, Australia) organizing committee of the HEp-2 contest with Gennaro Percannella (U. Salerno, Italy) Shiqi Yu representing the Alessia Saggese (U. Salerno, Italy) team from the Computer Mario Vento (U. Salerno, Italy) Vision Institute, Shenzhen University, Arnold Wiliem (U. Queensland, Australia) China, that won the competition.

This was the fourth of a series of were already proposed in previous submitted papers underwent benchmarking initiatives organized editions of the contest and used blind peer review in order to over the last few years and the same datasets in order to guarantee originality and technical hosted by the ICPR conference witness the advances made by soundness. (in 2012 and 2014) and by ICIP the scientific community on such Ten research teams coming from 2013, all of which attracted a tasks. Conversely, the remaining all the continents participated in very wide audience. The contest, two tasks were proposed to the the competition. The organizers that is a joint initiative by the community for the first time. received fifteen executables, five University of Salerno (Italy) and Following the experience for task 1, three for task 2, two for the University of Queensland of previous editions of the task 3 and five for task 4, and the (Australia) with the support of competition, the participants Moreover, the Program Committee Sullivan Nicolaides Pathology received a public set of data for carefully reviewed the paper (SNP), Australia, primarily aimed training their pattern recognition submissions, and six high-quality to provide a platform for scientists systems; then they had to submit papers were selected for oral and practitioners for performing the software implementation presentation and publication in the research to develop Computer of the proposed method in the ICPR2016 proceedings. Aided Diagnosis (CAD) systems form of an executable together for pathology tests utilizing indirect Results achieved on the private with all the parameterizations immunofluorescence protocol. test set were disclosed only during required to execute it; then, the the contest session held at ICPR The contest considered the organizers independently ran the 2016. Roughly two dozen people Antinuclear Antibodies (ANA) methods on the private test set. attended the session, having the test using Human Epithelial type Each research team was allowed opportunity to exchange ideas on 2 (HEp-2) cells. In particular, the to apply to any number of tasks the themes of the competition in a 2016 edition of the competition was with one executable with one very informal but productive way. divided into four tasks addressing set of parameters per task. The specific problems: (1) HEp-2 cell participants could also submit a For more information see: http:// classification; (2) Patient specimen regular paper in the ICPR 2016 mivia.unisa.it and http://www.itee. classification; (3) HEp-2 mitotic format describing their method uq.edu.au/sas/. cell identification and (4) Cell and the results achieved on the segmentation. The first two tasks public part of the dataset. The IAPR Newsletter, Vol. 39 No. 1, Feb. 2017 Page 37 Return to Highlights Return to Page 1 ICPR 2016 Highlights all the optimal solutions it’s hard Graph Distance Contest to tell how far its solutions are from the optimal ones and it is still https://gdc2016.greyc.fr/ not adapted to work with numeric attributes. DF and PDFS are An ICPR 2016 Contest, December 4, 2016, Cancún, Mexico highly dependent on the bipartite matching algorithm (used as a lower bound). One can notice that Scientific committee: there is still no method that can Luc Brun (Normandie University, ENSICAEN, France) match more than 40 nodes with Xiaoyi Jiang (University of Münster, Germany) really good deviation (compared to Jean-Yves Ramel (University of Tours, LI Tours, France) the optimal) Organizing committee: In future, more challenging Zeina Abu-Aisheh (University of Tours, LI Tours, France) datasets with all the optimal Sébastien Bougleux (Normandie University, UNICAEN, France) solutions, with bigger sizes of Benoit Gaüzère (Normandie Univ., INSA Rouen, LITIS, France) graphs, with denser graphs and with different types of attributes This contest focused on the Computation using a Binary have to be used. The influence of definition and the computation of Linear Program, Julien costs on the methods has also to general dissimilarity measures Lerouge, Zeina Abu-Aisheh, be studied more deeply. between graphs through two Romain Raveaux, Pierre challenges: Heroux, and Sebastien Adam, The goal of the Challenge 2 was, http://arxiv.org/pdf/1505.05740. given a dataset decomposed into a Challenge 1: computation of the pdf with 3 variants (24Threads) train set, a validation set and a test exact or an approximate graph edit • A Branch-and-Bound set, to determine the class of the distance algorithms: Zeina Abu-Aisheh, graphs in the test set with a usual Challenge 2: computation of a Romain Raveaux, Jean-Yves k-NN classifier (with k=3, decision dissimilarity measure for graph Ramel and Patrick Martineau at majority, reject if 3 different classification with 3 variants, http://www.rfai. classes). The evaluation was done li.univ-tours.fr/PagesPerso/ on four differents datasets Different datasets were considered zabuaisheh/anytimeGM.html with symbolic or numerical The two submitted methods were • Assignment Problem based analyzed and compared, for each attributes on both nodes and methods: Benoit Gaüzère, edges. dataset, according to recognition Sébastien Bougleux, Luc Brun, rate and confusion matrix. The first The goal of Challenge 1 was Vincenzo Carletti, Mario Vento method was High Order Stochastic to compute the exact or an https://hal.archives-ouvertes. Graphlets for Graph-Based Pattern approximate GED, for several fr/hal-01246709/file/technical_ Classification, Anjan Dutta and pairs of graphs of several datasets, report_ged.pdf with 2 variants Josep Lladós and the second one under the following constraints: The detailed results of this was the Upper Bound Depth-First running time limited to 30 seconds challenge show that F24threads, Graph Edit Distance Algorithm and edit costs imposed for each the method based on Binary Linear proposed by Zeina Abu-Aisheh, dataset. The submitted methods Program, is the most precise Romain Raveaux, Jean-Yves were compared according between exact methods but fails Ramel and Patrick Martineau. to running time closeness to to have a good deviation when reference distances. The reference The results demonstrate that the matching graphs whose sizes are both methods are interesting, distance is the optimal edit bigger than 60 vertices. QAPE distance (when available) or the as the first method wins on two beat F24threads on big graphs datasets (GREC, MAO) and the minimal distance found by the (graphs whose size is bigger than methods under evaluation. other one wins on Monoterp and 60 vertices). Indeed, QAPE was Mutagenicity datasets. The participants were: the most precise approximate • An Exact Graph Edit Distance algorithm but since we don’t have IAPR Newsletter, Vol. 39 No. 1, Feb. 2017 Page 38 Return to Highlights Return to Page 1 ICPR 2016 Highlights

MICHE-II Mobile Iris CHallenge Evaluation II

http://biplab.unisa.it/MICHE_Contest_ICPR2016/ An ICPR 2016 Contest, December 4, 2016, Cancún, Mexico

Organizers: Maria de Marsico (Sapienza University of Rome, Italy) Michele Nappi (University of Salerno, Italy) Hugo Proença (University of Beira Interior, Portugal) Modesto Castrillón Santana (University of Las Palmas de Gran Canaria, Spain)

Mobile biometric recognition by ICPR 2016, was to collect relevant ×2048 resolution, anterior personal and/or wearable devices contributions to the field of mobile FaceTime HD Camera with 1.2 provides a further application for iris recognition in both academy Megapixels (72 dpi) and 960 × user mobile equipment, which and industry. 1280$ resolution; are ubiquitous nowadays. On • Galaxy Tablet II (GT2), with In this new context, it is assumed the other hand, it extends the Google Android Operating that the subject to be recognized field of application of traditional System, no posterior camera, autonomously operates the biometric identification systems, 0.3 Megapixels anterior camera capturing device. MICHE-I by allowing capture of biometric with 640 × 480$ resolution. provided a dataset suitable traits in any place. The information to assess the performance of As a consequence, the three captured by mobile devices can be biometric applications related in groups of images have different compared with that stored either this specific set up. levels of resolution, which is one on the device itself, or even within of the factors that can negatively RFID tags, smartcards or machine The composition of the dataset affect cross-device recognition. readable identification documents used for MICHE-II challenge (IDs) for single user verification is basically the same as Image distortions in the MICHE- purposes. In addition, identification MICHE-I, with the addition of II dataset include: (a) reflections operations may be carried out on new unpublished images to be caused by light sources, people a remote server, for recognition in used mostly in each competitor's or objects in the scene; (b) out a set of relevant subjects. ranking process. Three kinds of of focus; (c) blur, due either to smartphones and tablets were an involuntary movement of the Iris is a natural candidate for used: hand holding the device, or to mobile biometric recognition for an involuntary movement of the two reasons: iris acquisition is little • a Galaxy Samsung IV (GS4), head/eye; (d) occlusions, due to intrusive, and iris codes are among with Google Android Operating eyelids, eyeglasses, eyelashes, the less expensive templates. Most System, CMOS posterior hair, or shadows; (e) device- current iris recognition systems camera with 13 Megapixel specific artifacts, due to the low still require that subjects stand (72dpi) and 2322 × 4128 resolution and/or to the specific close to the capture device (about resolution, CMOS anterior noise pattern of the device; 1m or less) and look towards it for camera with 2 Megapixel (f) off-axis gaze; (g)variable about 3s. The first iris biometric (72 dpi) and 1080 ×1920$ illumination; and (h) different color competitions have relied on images resolution; dominants. The lack of precise acquired in these conditions. The • iPhone5 (IP5), with Apple localization and of fixed distance aim of the contest Mobile Iris iOS Operating System, iSight in the capture causes both images CHallenge Evaluation II (MICHE- posterior camera with 8 containing well centred eyes and II), launched in conjunction with Megapixels (72 dpi) and 1536 images containing half faces to be IAPR Newsletter, Vol. 39 No. 1, Feb. 2017 Page 39 Return to Highlights Return to Page 1 ICPR 2016 Highlights present in dataset. This results in According to a policy established All selected approaches were variable sizes of the region useful by NICE competitions, the problem presented during the conference. for recognition. This is typical of of iris recognition was tackled A number of external conference mobile captures performed by the by two separate challenges: attendees, not participating to the user, which are usually neither MICHE-I dealt with the problem competition, were also present. too close nor at arm-length. This of segmentation of iris images Therefore a stimulating and introduces further difficulties, since acquired by mobile devices, and interesting discussion of methods eye localization must be performed the following MICHE-II started from and results was carried out, in a pre-processing step. In some the best segmentation algorithm which was appreciated by both cases, the resulting size of the as a fair preliminary processing organizers and participants. iris region is smaller. In other step to feed the following Details on the competition can cases, it is possible to exploit the phases of feature extraction and be found at: http://biplab.unisa. further possibilities offered by an classification. This algorithm was it/MICHE_Contest_ICPR2016/ extended periocular region. The provided to all competing groups and results at http://biplab.unisa. dataset has been collected during registered for the challenge, it/MICHE_Contest_ICPR2016/ several different data acquisition in order to get an unbiased finalrank.php. Presentations at the sessions separated in time. The comparison. conference can be found at: http:// time elapsed between the first and Out of the 12 submitted algorithms, biplab.unisa.it/MICHE_Contest_ second acquisition of a subject six were selected to be presented ICPR2016/presentations.php. varies from a minimum of two at ICPR2016, namely the ones months to a maximum of nine. At achieving the best performance. present, MICHE-I contains images Moreover, the authors were invited from 75 different subjects, with to submit an extended version of 1297 by GS4, 1262 images from their work to a special issue of IP5, and 632 images from GT2. Pattern Recognition Letters.

Some of the session participants

IAPR Newsletter, Vol. 39 No. 1, Feb. 2017 Page 40 Return to Highlights Return to Page 1 ICPR 2016 Highlights

AcTiVComp16 Contest on Arabic Text Detection and Recognition in Video Frames

An ICPR 2016 Contest, December 4, 2016, Cancún, Mexico

Organizers: Oussama Zayene (Tunisia), Nadia Hajjej (Tunisia), Soumaya Ben Mansour (Tunisia) Sameh Masmoudi Touj (Tunisia), Jean Hennebert (Switzerland), Rolf Ingold (Switzerland), and Najoua Essoukri Ben Amara (Tunisia)

Introduction: embedded Arabic text in videos. of participating systems to To this end, we have organized automatically locate and In the last two decades, important the first Arabic video text detection recognize overlay texts using progress has been achieved in and recognition competition— our freely available AcTiV1 the conventional field of Arabic AcTiVComp2 in the context of ICPR dataset and OCR (e.g., scanned documents) 2016. in terms of systems, datasets 2. to unify the groundtruth and competitions. However, few The main objectives of this contest formats, the evaluation attempts have been made on were protocols and metrics used in the development of detection the Arabic Video OCR domain. 1. to evaluate the performance and recognition systems for

Fig1. Typical video frames from AcTiV dataset. From left to right examples of RussiaToday, France24, TunisiaNat1 and AljazeeraHD channels AcTiVComp16: stability filtering and heuristic Ohyama, from the Graduate merging. School of Engineering, In this first edition of AcTiVComp • FM-AVTD system was Mie University, Japan. This four groups participated with five submitted by Yang Xue-hang system is based on projection systems: from the NLPR “National profile analysis, Difference of Three systems for the detection Laboratory of Pattern Gaussians (DOG) filter and task: Recognition”, Institute of eccentricity reduction. Automation, Chinese Academy Two systems for the recognition • ATD-CH system was submitted of Sciences, China. This task: by Houda Gaddour from system consists of MSER the MIRACL “Multimedia, algorithm, AdaBoost classifier • ATR-SID system was submitted InfoRmation systems and and rule-based post-processing by Soumaya Essefi from the Advanced Computing module. National Engineering School of Laboratory”, University of • D-ATVF system was submitted Sousse, University of Sousse Sfax, Tunisia. This system is by Seiya Iwata and Wataru Tunisia. This system is based based on color clustering, area IAPR Newsletter, Vol. 39 No. 1, Feb. 2017 Page 41 Return to Highlights Return to Page 1 ICPR 2016 Highlights

on Multi Dimensional Recurrent protocols (six for each task) were ATR-SID system for “France24”, Neural Networks (MDRNNs) used to evaluate the participating “RussiaToday” and “TunisiaNat1” coupled with Connectionist systems under the same channel-dependent protocols and Temporal Classification (CTC). experimental conditions. These by the R-AVTF system for protocol protocols can be divided into 6.4. The contest results show that • R-ATVF system was submitted two groups: channel-dependent there is still room for improvement by the authors of the D-ATVF protocols e.g., protocol 4.3 for in both detection and recognition system. This system is TunisiaNat1 TV channel (detection of Arabic video text. We look based on modified quadratic task) and cross-channels (or forward to having more participants discriminant function (MQDF) channel-free) protocols e.g. in future editions of AcTiVComp and dynamic programming protocol 6.4 for All-SD channels and more researchers joining (DP). (recognition task). More details the challenging research topic of The performance of detection about the protocols are explained Arabic video text detection and systems was measured using the in the contest paper2. recognition. standard evaluation metrics i.e. In the detection task, the best recall, precision and f-score using 1. O. Zayene, J. Hennebert, S. M. Touj, R. results were yielded by the FM- our open-source evaluation tool. Ingold, and N. Essoukri Ben Amara, “A AVTD system for “AljazeeraHD”, The recognition systems were dataset for Arabic text detection, tracking “France24” and “TunisiaNat1” evaluated based on the recognition and recognition in news videos- AcTiV”, channel-dependent protocols and in Proc. of ICDAR'15, Nancy- France, rates at the character, word and “All-SD” cross-channels protocol August 2015. line levels i.e. CRR, WRR and 2 and by the D-ATVF system for Zayene O., Hajjej N., Touj S. M., Ben LRR. The participating systems “RussiaToday” channel-dependent Mansour S., Hennebert J., Ingold R. et were tested in a blind manner protocol. Ben Amara N. E., “ICPR2016 Contest on on the closed-set of the AcTiV Arabic Text Detection and Recognition in dataset which was unknown to all In the recognition task, the best Video Frames—AcTiVComp”, ICPR’16, participants. Twelve evaluation results were yielded by the Cancún Mexico 2016. AcTiVComp17

http://diuf.unifr.ch/diva/AcTiVComp/ will take place in conjunction with ICDAR2017 (November 10-15, 2017, Kyoto, Japan)

Video-OCR system pipeline Video-OCR context, challenges and objectives

IAPR Newsletter, Vol. 39 No. 1, Feb. 2017 Page 42 Return to Highlights Return to Page 1 ICPR 2016 Highlights 2016 IAPR Fellows Photo by Rangachar Kasturi Rangachar by Photo Mohammad Showkat-Ul Alam (USA) Tao Mei (China) For contributions to the advancement and development of For contributions to large-scale video analysis, understanding, and architectures, algorithms, and performance metrics for optical applications pattern recognition and high resolution image reconstruction Sushmita Mitra (India) Richard Bowden (UK) For contributions to neuro-fuzzy and hybrid approaches to For contributions to computer vision in the fields of sign language, pattern recognition and machine intelligence, with applications to gesture, and activity recognition and for service to IAPR bioinformatics and medical imaging

Alberto Broggi (Italy) Maja Pantic (UK) For contributions in the field of computer vision-based For contributions to affective and behavioural computing advanced driver assistance systems and autonomous driving Petia Radeva (Spain) Xilin Chen (China) For contributions to computer vision, machine learning, egocentric For contributions to image modeling and object recognition vision, life-logging, and medical imaging

Mohamed Cheriet (Canada) Amit K Roy-Chowdhury (USA) For contributions to ancient manuscript processing, For contributions to collaborative sensing and distributed form construction, handwriting recognition, processing in camera networks with applications in tracking, multidisciplinary application, and for service to IAPR re-identification, and activity recognition

Aly A Farag (USA) Punam Kumar Saha (USA) For contributions to object modeling and biomedical applications For contributions to digital topology and geometry and their application Yun Fu (USA) For contributions to pattern recognition, data mining, Stan Sclaroff (USA) and visual intelligence For contributions in tracking, human gesture analysis, shape recognition, and video databases Anders Lars-Gunnar Heyden (Sweden) For contributions to multiple view geometry and auto-calibration Zhenan Sun (China) and for service to IAPR For contributions to automatic iris recognition and applications

Gang Hua (USA) Emanuele Trucco (UK) For contributions to visual computing and learning from For contributions to computer vision and applications unconstrained images and videos René Vidal (USA) For contributions to computer vision and pattern recognition C. V. Jawahar (India) For contributions to computer vision, pattern recognition, and Dong Xu (Australia) document image understanding For contributions to machine learning for visual recognition and domain adaption Kenichi Kanatani (Japan) For contributions to 3D computer vision analysis and computation Jian Yang (China) For contributions to 2D image representation Ajay Kumar (Hong Kong) and discriminant analysis For contributions to human identification using biometrics

Zhouchen Lin (China) Naokazu Yokoya (Japan) For contributions to image processing, computer vision, For contributions to image processing, computer vision and and mixed and augmented reality machine learning

Derong Liu (China) Jie Zhou (China) For contributions to biometrics, computer vision, For contributions to neural networks and associative memories and image processing

IAPR Newsletter, Vol. 39 No. 1, Feb. 2017 Page 43 Return to Highlights Return to Page 1 ICPR 2016 Highlights

tions, ICPR 2016 Award gratula Recipi Con ents!

ICPR 2016 Best Paper Awards

Best Industry Related Paper Award presented to Niki Martinel, Gian Luca Foresti, and Christian Micheloni for the 23rd ICPR Paper "Distributed and Unsupervised Cost-Driven Person Re-Identification"

Piero Zamperoni Best Student Paper Award presented to Sungmin Eum for the 23rd ICPR Paper "Content Selection Using Frontalness Evaluation of Multiple Frames" authors: Sungmin Eum and David Doermann

NOTE: Sungmin Eum is the author of the IAPR...The Next Generation column in this issue of the IAPR newsletter. ~A. Kuijper, EiC, IAPR Newsletter

Photos: Diego Sánchez/Victor Alvarez

IAPR Newsletter, Vol. 39 No. 1, Feb. 2017 Page 44 Return to Highlights Return to Page 1 ICPR 2016 Highlights

Intel Best Scientific Paper Awards

Track 1: Pattern Recognition and Machine Learning Qing Tian and Songcan Chen "Cross-Heterogeneous-Database Age Estimation with Co-Representation among Them"

Track 2: Computer Vision and Robot Vision Susanna Glad, Martin Danelljan, Fahad Shahbaz Khan, and Michael Felsberg "Deep Motion Features for Visual Tracking"

Track 3: Image, Speech, Signal and Video Processing Dan Xu, Jingkuan Song, Xavier Alameda-Pineda, Elisa Ricci, and Nicu Sebe "Multi-Paced Dictionary Learning for Cross-Domain Retrieval and Recognition"

Track 4: Document Analysis, Biometrics and Pattern Recognition Applications Yusuf Osmanlioglu, Santiago Ontañón, Uri Hershberg, and Ali Shokoufandeh "Efficient Approximation of Labeling Problems with Applications to Immune Repertoire Analysis"

Track 5: Biomedical Image Analysis and Applications Serge Bobbia, Yannick Benezeth, and Julien Dubois "Remote Photoplethysmography Based on Implicit Living Skin Tissue Segmentation"

Photo: Diego Sánchez/Victor Alvarez

IAPR Newsletter, Vol. 39 No. 1, Feb. 2017 Page 45 Return to Highlights Return to Page 1 ICPR 2016 Highlights

IBM Best Student Paper Awards

Track 1: Pattern Recognition and Machine Learning Yangwei Liu "One-Pass Online SVM with Extremely Small Space Complexity" authors: Yangwei Liu and Jinhui Xu

Track 2: Computer Vision and Robot Vision Satoshi Morinaka "3D Reconstruction under Light Ray Distortion from Parametric Focal Cameras" authors: Satoshi Morinaka, Fumihiko Sakaue, Jun Sato, Kazuhisa Ishimaru, and Naoki Kawasaki

Track 3: Image, Speech, Signal and Video Processing Bo Wang "Adaptive Boosting for Image Denoising: Beyond Low-Rank Representation and Sparse Coding" authors: Bo Wang, Tao Lu, and Zixiang Xiong

Track 4: Document Analysis, Biometrics and Pattern Recognition Applications Andras Rozsa "Are Facial Attributes Adversarially Robust?" authors: Andras Rozsa, Manuel Günther, Ethan Rudd, and Terry Boult

Track 5: Biomedical Image Analysis and Applications Jennifer Alvén "Shape-Aware Multi-Atlas Segmentation" authors: Jennifer Alvén, Fredrik Kahl, Matilda Landgren, Viktor Larsson, and Johannes Ulén

Photo: Diego Sánchez/Victor Alvarez

IAPR Newsletter, Vol. 39 No. 1, Feb. 2017 Page 46 Return to Highlights Return to Page 1 ICPR 2016 Highlights

IAPR Service Awards

IAPR/ICPR 2016 Distinguished Performance Award presented to Prof. Brian Carrington Lovell in recognition of outstanding performance as Program Chair of the 23rd International Conference on Pattern Recognition

IAPR Certificate of Appreciation presented to Linda O'Gorman for highly professional support to the IAPR ExCo and for service to the IAPR

Photos: Joakim Nyström and Diego Sánchez/Victor Alvarez

IAPR Newsletter, Vol. 39 No. 1, Feb. 2017 Page 47 Return to Highlights Return to Page 1 ICPR 2016 Highlights

Elsevier Awards presented at ICPR 2016

Elsevier Pattern Recognition Journal Award of Excellence presented to Professor Ching Y. Suen in recognition of his distinguished editorial services (1984-2016)

Elsevier Journal of the Pattern Recognition Society Pattern Recognition Best Paper Award for 2013 presented to Zhiquan Qi, Yingjie Tian, Yong Shi received by Mohamed Cheriet (ETS, Canada) in place of authors for the paper entitled "Robust twin support vector machine for pattern classification" Pattern Recognition. Volume 46, Issue 1, Jan. 2013, Pages 305-316.

Elsevier Journal of the Pattern Recognition Society Pattern Recognition Best Paper Award for 2014 presented to Asja Fischer and Christian Igel receivedby Asja Fischer for the paper entitled "Training restricted Boltzmann machines: An introduction" Pattern Recognition. Volume 47, Issue 1, Jan. 2014, Pages 25-39.

Photos: Diego Sánchez/Victor Alvarez

IAPR Newsletter, Vol. 39 No. 1, Feb. 2017 Page 48 Return to Highlights Return to Page 1 ICPR 2016 Highlights The 2016 Meeting of the IAPR Governing Board

The IAPR Governing Board meets every two years during the ICPR conference. Representatives from all of the IAPR's member societies get together to discuss and vote on matters of high importance to the governance of the IAPR. Some of the key outcomes of the 2016 Governing Board meeting are listed below. (See the IAPR News at the IAPR website and future issues of the IAPR Newsletter for more information as it becomes available.)

• New member societies from Vietnam and Pakistan were approved by a unanimous GB vote. • The venue of ICPR 2020 will be Milan, Italy. This venue was selected by vote from among five competing bids of excellent quality. • An increase in annual membership dues was approved. Annual dues had remained the same for the past 20 years. The previous good financial status of IAPR had enabled new initiatives, such as TC Summer Schools, Research Scholarships, and IAPR Travel stipends. In order to sustain these initiatives, the IAPR needs to begin implementing a new financial plan.The modest dues increase is part of that plan.

IAPR Executive Committee for the 2016-2018 Term

Photo: Diego Sánchez/Victor Alvarez The New IAPR ExCo: Secretary Alexandra Branzan-Albu, Treasurer Apostolos Antonacopoulos, President Simone Marinai, First Vice-President Massimo Tistarelli, Second Vice-President Edwin Hancock, Past President Ingela Nyström

IAPR Newsletter, Vol. 39 No. 1, Feb. 2017 Page 49 Return to Highlights Return to Page 1 This bulletin board contains items of interest to the IAPR Community

APPLICATION DEADLINE: 31 MARCH 2017

The 11th edition of the International Computer Vision Summer School aims to provide both an objective and clear overview and an in-depth analysis of the state-of-the-art research in Computer Vision. DISCOUNT

click anywhere on this notice to go to: http://iplab.dmi.unict.it/icvss2017/

The IAPR and Springer Clink on links below so you don't miss these important items in have a new agreement that allows IAPR members this issue: to get Letter from the New discounted subscriptions President of the IAPR to the International Journal on Calls for Papers Document Analysis and Recognition (IJDAR) and Machine Vision and Calls from IAPR Committees Applications (MVA).

Contact Linda O'Gorman, [email protected] for more information

IAPR Newsletter, Vol. 39 No. 1, Feb. 2017 Page 50 Return to Page 1 Meeting and Education Planner The IAPR web site has the most up-to-date information on IAPR events. Click here. NOTE: Highlighting indicates that the paper submission deadline is still open. * Asterisks denote non-IAPR events * Report on Meeting previous Venue edition ICPRAM 2017: 6th International Conference on Portugal

FEB Pattern Recognition Applications and Methods CCIW 2017: 6th Computational Color Imaging Workshop CCIW 2015 Italy MAR ASAR 2017: First Intl. Workshop on Arabic Script Analysis and Recognition France

APR IWBF 2017: 5th International Workshop on Biometrics and Forensics IWBF 2016 UK MVA 2017: 15th IAPR Intl. Conf. on Machine Vision Applications MVA 2015 Japan GbR 2017: 11th IAPR-TC15 Workshop on GbR 2015 Italy MAY Graph-based Representations SCIA 2017: 20th Scandinavian Conference on Image Analysis SCIA 2015 Norway Biometrics SSB 2017: 14th IAPR/IEEE International Summer School on Biometrics Italy 2016 IGS 2017: 18th International Graphonomics Society Conference Italy

JUN MCPR 2017: 9th Mexican Congress on Pattern Recognition MCPR 2016 Mexico * ICVSS 2017: International Computer Vision Summer School: 2017 ICVSS 2016 Italy From Representation to Action and Interaction *

JUL ICPRS 2017: 8th International Conference on Pattern Recognition Systems Spain * CAIP 2017: 17th International Conference on Computer Analysis of Images Sweden AUG and Patterns * * ICIAP 2017: 19th Intl. Conference on Image Analysis and Processing * ICIAP 2015 Italy * GCPR 2017: 39th German Conference on Pattern Recognition * Germany DGCI 2017: 20th International Conference on Discrete Geometry for DGCI 2016 Austria

SEP Computer Imagery CIARP 2017: 22nd Iberoamerican Congress on Pattern Recognition Chile ICDAR 2017: 14th International Conference on Japan Document Analysis and Recognition

PSIVT 2017: 8th Pacific Rim Symposium on Image and Video Technology PSIVT 2015 China NOV PreMI 2017: 7th International Conference on Pattern Recognition and PreMI 2015 India

DEC Machine Intelligence

The IAPR Newsletter is published in association with the IAPR website, www.iapr.org The IAPR Newsletter is published four times per year, January, April, July, and October.

Deadline for the next issue: March 27, 2017 To contact us: Arjan Kuijper, Editor-in-Chief, [email protected] Owais Mehmood, Associate Editor for Book Reviews, [email protected] Linda J. O'Gorman, Layout Editor, [email protected]

IAPR Newsletter, Vol. 39 No. 1, Feb. 2017 Page 51 Return to Page 1