An Overview of the Weocr System and a Survey of Its Use’, Proceedings of Image and Vision Computing New Zealand 2007, Pp

An Overview of the Weocr System and a Survey of Its Use’, Proceedings of Image and Vision Computing New Zealand 2007, Pp

H. Goto, ‘An Overview of the WeOCR System and a Survey of its Use’, Proceedings of Image and Vision Computing New Zealand 2007, pp. 121–125, Hamilton, New Zealand, December 2007. An Overview of the WeOCR System and a Survey of its Use Hideaki Goto Information Synergy Center, Tohoku University, Sendai, 980–8578 Japan. http://www.sc.isc.tohoku.ac.jp/∼hgot/ email: [email protected] Abstract The collaboration of the network and optical character recognition (OCR) is expected to be useful for extending the applications of OCR, and it has a potential to open up new vistas in future services using OCR. We presented the basic ideas of the synergetic OCR system in 2004, and developed a platform called WeOCR to realize the web-based OCR services. The number of accesses has been increasing gradually since the interoperable, synergetic web-based OCR system was put into service in 2005. In this paper, we describe an overview of the current WeOCR system and provide a survey of its use. The unique survey may convey some useful hints for future development of network-oriented OCRs, not limited to the WeOCR, and potential applications of such systems. Keywords: web-based OCR, WeOCR, OCRGrid, character recognition, document analysis 1 Introduction tem has become very difficult for a researcher or a small group of people. A possible solution would The collaboration of the network and optical char- be to share software components as Free Software acter recognition (OCR) is expected to be useful among researchers and developers. However, many for making lightweight applications not only for researchers think the programs are a kind of in- desktop systems but also for small gadgets and tellectual property and are not willing to release autonomous robots. the source codes even after the related papers have For example, some mobile phones with character been published. An alternative solution is to use recognition capabilities are available in the mar- Web Application Servers. Since the programs run ket today [1]. A mobile phone with a real-time on the server side, people can provide comput- English–Japanese translator was proposed by a ing (pattern recognition) power to others without Japanese company. Although it has a simple, built- having their source codes or executables open. A in OCR, it requires a large language dictionary prototype system for a text locating competition provided by an external server on the Internet. An- was proposed [3]. other example is the vision system of autonomous In 2004, we presented a new concept of network- robots. We have been working on an autonomous oriented OCR systems called Synergetic OCR [4]. robot system with text detection and character The basic ideas are totally different from those of recognition capabilities [2]. Due to the limitation some experimental websites with OCR capability of the hardware, it is difficult to put a large character- that existed at that time for demonstration pur- set data and a dictionary for language process- poses. We have designed the Synergetic OCR to ing into small gadgets or robots. In addition, we make a lot of OCR engines work cooperatively over cannot use a sophisticated character recognition the network worldwide to gain some synergetic ef- method, because the processor power is very lim- fects such as performance improvement of OCR, ited. Thus, use of network can be beneficial to to realize multilingual, high-function, sophisticated various applications. OCR systems, and to provide ubiquitous OCR ser- The network-based architecture has some advan- vices. We have revised the Synergetic OCR later tages from the researchers’ and developers’ points and proposed a new conceptual platform called of view as well. Recent OCR systems are becom- OCRGrid in 2006 [5]. ing more and more complicated, and the devel- Since year 2004, we have been designing a web- opment requires expertise in various fields of re- based OCR system, called WeOCR, as a simplified searches. Building and studying a complete sys- instance of the OCRGrid. The WeOCR system 121 cooperative work OCR server OCR OCR server server OCR OCR server server send images OCR server client receive recognition results client WAN/LAN Figure 1: OCRGrid platform. Figure 2: Directory-based server search system. was put into service in 2005. One of the biggest concerns of a web-based OCR is about the privacy 2.2 WeOCR Toolkit of documents. A simple question arises — Do peo- Making a web-based OCR from scratch as an ap- ple really want to use a system like that? To answer plication server requires a lot of expertise about this question, we probably need to investigate the network programming and network security. Since usage of the system. many OCR developers and researchers are not so familiar with network programming, we have de- In this paper, we describe an overview of the cur- 1 rent WeOCR system, present the usage statistics, veloped a toolkit to help those people build secure and provide a survey of its use to obtain some web-based OCRs easily. The toolkit consists of useful hints for future development of the WeOCR some interface programs and filter programs that and potential applications of OCR. work together with the web server. The document image is sent from the client com- puter to the WeOCR toolkit via the Apache web 2 Overview of the WeOCR Sys- server program. The programs in the toolkit check tem the image size, examine the file integrity, uncom- press the file if necessary, convert the image file into 2.1 Basic Concept a common file format, and invoke the OCR soft- ware. The toolkit includes some programs used for The OCRGrid is a conceptual platform, while the protecting the web server from malicious attacks WeOCR is an implementation of the OCRGrid based or from the defects of the OCR programs. on HTTP (HyperText Transfer Protocol). The recognition results are converted into HTML The basic concept of the OCRGrid is rather simple. data and sent back to the client. A lot of OCR servers are deployed on a network as shown in Figure 1. The OCR servers work To enable the end users to search for appropriate either independently or cooperatively communicat- OCR engines easily, we have developed a directory- ing with each other. A client connects to one of based server search system. Figure 2 depicts the the OCR servers and sends text images to it. The overview. servers recognize the images and send the recogni- A specification file written by the server adminis- tion results (i.e. character codes, etc.) back to the trator is attached to each OCR engine. The speci- client. fication file is written in XML (eXtensible Markup Although the OCRGrid is expected to be used world- Language) so that automated data handling be- wide over the Internet, we may use any networks comes easy. The file describes the specifications such as wireless networks, corporate local area net- of the server, including the location (URL) of the works (LANs), virtual private networks (VPNs), server, the location of the application interface pro- interconnection networks in parallel servers, and gram (CGI), the name of the OCR engine used, even interprocess communication channels. supported languages, supported document types, etc. The portal server has a robot program for Some potential applications of the platform have collecting the specification data automatically and been discussed in [4, 5]. periodically from the OCR servers. The robot ana- Unlike some commercial OCR systems equipped lyzes each specification file in XML and update the with web interfaces, we are expecting the WeOCR database entries. A simple search program picks servers will be supported by the communities of up the OCR servers that match the client’s needs researchers, developers, and individuals as well as from the database and shows the search results. application service providers. The toolkit and the server search system may be easily adapted to other online applications related 1The toolkit is available at http://weocr.ocrgrid.org/ 122 Table 1: WeOCR servers (as of Aug. 2007) Server (location) Sub-ID Engine Deployed Description ocr1.sc ocrad GNU Ocrad Feb 2005 Western European languages (Japan) gocr GOCR Nov 2005 Western European languages appsv.ocrgrid ocrad GNU Ocrad Jun 2006 Western European languages (Japan) hocr HOCR Jun 2006 Hebrew OCR tesseract Tesseract OCR Aug 2006 English OCR scene Tesseract OCR Apr 2007 Scene text recognition for English + private preprocessor (experimental) ocr.ekitap ocrad GNU Ocrad Aug 2006 Turkish OCR (Turkey) + private postprocessor asv.aso ocrad GNU Ocrad Aug 2006 Western European languages (Japan) tesseract Tesseract OCR Aug 2006 English OCR to computer vision and pattern recognition. the two WeOCR servers, appsv.ocrgrid and ocr1.sc, in our laboratory. The numbers of requests to 2.3 Available Servers six engines are added together. The line “access” represents the numbers of requests, while another Table 1 shows the WeOCR servers available on the one “image ok” shows the numbers of successful Internet as of Aug. 2007. image uploading. The failures are mainly due to All the servers are constructed based on Open Source file format mismatch and interrupts of data trans- OCR software so far. Although these OCR engines missions. work very well with some limited kinds of docu- The number of accesses has been increasing grad- ments, the average performance is basically inferior ually since the first server was put into service in to that of the commercial products. Thanks to the 2005, despite the privacy concerns. We have about Tesseract OCR [6], the situation is improving. The 3,000 accesses per month today. This suggests that Tesseract OCR, under intensive development at there are a lot of applications of OCR in which Google, outperforms other Open Source OCR en- privacy does not matter.

View Full Text

Details

  • File Type
    pdf
  • Upload Time
    -
  • Content Languages
    English
  • Upload User
    Anonymous/Not logged-in
  • File Pages
    5 Page
  • File Size
    -

Download

Channel Download Status
Express Download Enable

Copyright

We respect the copyrights and intellectual property rights of all users. All uploaded documents are either original works of the uploader or authorized works of the rightful owners.

  • Not to be reproduced or distributed without explicit permission.
  • Not used for commercial purposes outside of approved use cases.
  • Not used to infringe on the rights of the original creators.
  • If you believe any content infringes your copyright, please contact us immediately.

Support

For help with questions, suggestions, or problems, please contact us