DOCSLIB.ORG

Arxiv:1901.02840V2 [Cs.CV] 8 Apr 2019 and Small file Sizes

Total Page：16

File Type：pdf, Size：1020Kb

Download full-text PDF Read full-text
Arxiv:1901.02840V2 [Cs.CV] 8 Apr 2019 and Small file Sizes GIF2Video: Color Dequantization and Temporal Interpolation of GIF images Yang Wang1, Haibin Huang2y, Chuan Wang2, Tong He3, Jue Wang2, Minh Hoai1 1Stony Brook University, 2Megvii Research USA, 3UCLA, yCorresponding Author Abstract Color Color Quantization Dithering Graphics Interchange Format (GIF) is a highly portable graphics format that is ubiquitous on the Internet. De- spite their small sizes, GIF images often contain undesir- !"#"$ %&#'((' )$$"$ *+,,-.+"/ able visual artifacts such as flat color regions, false con- tours, color shift, and dotted patterns. In this paper, we propose GIF2Video, the first learning-based method for en- hancing the visual quality of GIFs in the wild. We focus Artifacts: Artifacts: 1. False Contour 4. Dotted Pattern on the challenging task of GIF restoration by recovering 2. Flat Region information lost in the three steps of GIF creation: frame 3. Color Shift sampling, color quantization, and color dithering. We first Figure 1. Color quantization and color dithering. Two major propose a novel CNN architecture for color dequantization. steps in the creation of a GIF image. These are lossy compression It is built upon a compositional architecture for multi-step processes that result in undesirable visual artifacts. Our approach is able to remove these artifacts and produce a much more natural color correction, with a comprehensive loss function de- image. signed to handle large quantization errors. We then adapt the SuperSlomo network for temporal interpolation of GIF frames. We introduce two large datasets, namely GIF-Faces and GIF-Moments, for both training and evaluation. Ex- temporal resolution of the image sequence by using a mod- perimental results show that our method can significantly ified SuperSlomo [20] network for temporal interpolation. improve the visual quality of GIFs, and outperforms direct The main effort of this work is to develop a method for baseline and state-of-the-art approaches. color dequantization, i.e., removing the visual artifacts in- troduced by heavy color quantization. Color quantization is a lossy compression process that remaps original pixel col- 1. Introduction ors to a limited set of entries in a small color palette. This GIFs [1] are everywhere, being created and consumed process introduces quantization artifacts, similar to those by millions of Internet users every day on the Internet. The observed when the bit depth of an image is reduced. For 48 widespread of GIFs can be attributed to its high portability example, when the image bit depth is reduced from -bit to 24-bit, the size of the color palette shrinks from 2:8×1014 arXiv:1901.02840v2 [cs.CV] 8 Apr 2019 and small file sizes. However, due to heavy quantization 1:7 × 107 in the creation process, GIFs often have much worse visual colors to colors, leading to a small amount of ar- quality than their original source videos. Creating an ani- tifacts. The color quantization process for GIF, however, mated GIF from a video involves three major steps: frame is far more aggressive with a typical palette of 256 dis- sampling, color quantization, and optional color dithering. tinct colors or less. Our task is to perform dequantization Frame sampling introduces jerky motion, while color quan- from a tiny color palette (e.g., 256 or 32 colors), and it is tization and color dithering create flat color regions, false much more challenging than traditional bit depth enhance- contours, color shift, and dotted patterns, as shown in Fig.1. ment [15, 25, 38]. In this paper, we propose GIF2Video, the first learning- Of course, recovering all original pixel colors from the based method for enhancing the visual quality of GIFs. Our quantized image is nearly impossible, thus our goal is to algorithm consists of two components. First, it performs render a plausible version of what the original image might color dequantization for each frame of the animated gif look like. The idea is to collect training data and train a sequence, removing the artifacts introduced by both color ConvNet [16, 23, 29, 34, 44] to map a quantized image to quantization and color dithering. Second, it increases the its original version. It is however difficult to obtain a good 1 dequantization network for a wide range of GIF images. tion of the input GIFs. To this end, we propose two novel techniques to improve the performance of the dequantization network. Firstly, 3. GIF Generation and Artifacts we pose dequantization as an optimization problem, and we propose Compositional Color Dequantization Network The three main steps of creating a GIF from a video are: (CCDNet), a novel network architecture for iterative color (1) frame sampling, (2) color quantization, and (3) color dequantization. Similar to the iterative Lucas-Kanade algo- dithering. Frame sampling reduces the file size of the ob- rithm [28], this iterative procedure alleviates the problems tained GIF, but it also lowers the temporal resolution of the associated with severe color quantization. Secondly, during video content. In this section, we will provide more details training, we consider reconstruction loss and generative ad- about the color quantization and color dithering processes versarial loss [10, 19, 32] on both pixel colors and image and the resulting visual artifacts as seen in Figure1. gradients. This turns out to be far more effective than a loss 3.1. GIF Color Quantization function defined on the color values only. Another contribution of the paper is the creation of two The GIF color quantization process takes an input im- large datasets: GIF-Faces and GIF-Moments. Both datasets age I 2 RH×W ×3 and a color palette C 2 RN×3 of N contain animated GIFs and their corresponding high-quality distinct colors, and produces a color-quantized GIF image videos. GIF-Faces is face-centric whereas GIF-Moments G. The quantization is computed for each pixel, thus G has is more generic and diverse. Experiments on these two the same width and height as the input image. Gi;j at pixel datasets demonstrate that our method can significantly en- (i; j) is simply set to the color in the palette C closest to hance the visual quality of GIFs and reduce all types of ar- 2 the input color Ii;j, i.e., Gi;j = argminc 2 C Ii;j − c 2. tifacts. Comparatively, Our method outperforms its direct The color palette C could be optimized with a clustering al- baselines as well as existing methods such as False Contour 2 gorithm to minimize the total quantization errorkI − Gk2. Detection & Removal [15] and Pix2Pix [19]. Different clustering algorithms are used in practice, but Me- dian Cut [13] is the most popular one due to its computa- 2. Related Work tional efficiency. Most of the visual artifacts in GIFs are produced by the False Contour Detection and Removal. Smooth areas of color quantization process with a tiny color palette (N = images and video frames should not contain color edges, 256; 32; :::). As illustrated in Figure1, the three most no- but false contours are often visible in those areas after color ticeable types of artifacts are (1) flat regions, (2) false con- bit-depth reduction or video codec encoding. Several false tours, and (3) color shift. We notice a GIF image has a lot contour detection and decontouring methods [2,5,7, 15, 21, of connected components in which the color values are the 24, 42] have been proposed to address this problem. Among same, which will be referred to as “flat regions”. Flat re- them, False Contour Detection and Removal (FCDR) [15] gions are created because neighboring pixels with similar is the most recent state-of-the-art approach. It first locates colors are quantized into the same color bin in the palette. the precise positions of the false contours and then applies False contours also emerge at the boundaries between flat dedicated operations to suppress them. However, the color regions with close color values. This is because the con- quantization artifacts in GIFs are far more severe, and GIF tinuity of the color space has been broken and the color color dequantization requires more than removing minor change cannot be gradual. We also notice the color shift false contours produced by bit-depth reduction. between the input image and the GIF is larger for certain Video Interpolation. Classical video interpolation meth- small regions such as the lips of the baby in Figure1. This ods rely on cross-frame motion estimation and occlusion is because the color palette does not spend budget on these reasoning [3,4, 14, 18]. However, motion boundaries and small regions even though they have unique, distinct colors. severe occlusions are still challenging for existing optical 3.2. GIF Color Dithering flow estimation methods [6,9]. Moreover, the flow com- putation, occlusion reasoning, and frame interpolation are Color quantization with a small color palette yields sub- separated steps that are not properly coupled. Drawing in- stantial quantization error and artifacts. Color dithering is spiration from the success of deep learning in high-level vi- a technique that can be used to hide quantization error and sion tasks [12, 22, 36], many deep models have been pro- alleviate large-scale visual patterns such as false contours in posed for single-frame interpolation [26, 27, 30, 31] and GIFs. The most popular color dithering approach is Floyd- multi-frame interpolation [20]. SuperSlomo [20] is a re- Steinberg dithering [8]. It diffuses the quantization error cently proposed state-of-the-art method for variable-length from every pixel to its neighboring pixels in a feedback multi-frame interpolation approach. We adapt this method process. The dithered GIF has the same exact small color for GIF frame interpolation to enhance the temporal resolu- palette. However, it appears to have more colors.
Load more

Recommended publications
  • Snakes, Shapes, and Gradient Vector Flow
    IEEE TRANSACTIONS ON IMAGE PROCESSING, VOL. 7, NO. 3, MARCH 1998 359 Snakes, Shapes, and Gradient Vector Flow Chenyang Xu, Student Member, IEEE, and Jerry L. Prince, Senior Member, IEEE Abstract—Snakes, or active contours, are used extensively in There are two key difficulties with parametric active contour computer vision and image processing applications, particularly algorithms. First, the initial contour must, in general, be to locate object boundaries. Problems associated with initializa- close to the true boundary or else it will likely converge tion and poor convergence to boundary concavities, however, have limited their utility. This paper presents a new external force to the wrong result. Several methods have been proposed to for active contours, largely solving both problems. This external address this problem including multiresolution methods [11], force, which we call gradient vector flow (GVF), is computed pressure forces [10], and distance potentials [12]. The basic as a diffusion of the gradient vectors of a gray-level or binary idea is to increase the capture range of the external force edge map derived from the image. It differs fundamentally from fields and to guide the contour toward the desired boundary. traditional snake external forces in that it cannot be written as the negative gradient of a potential function, and the corresponding The second problem is that active contours have difficulties snake is formulated directly from a force balance condition rather progressing into boundary concavities [13], [14]. There is no than a variational formulation. Using several two-dimensional satisfactory solution to this problem, although pressure forces (2-D) examples and one three-dimensional (3-D) example, we [10], control points [13], domain-adaptivity [15], directional show that GVF has a large capture range and is able to move snakes into boundary concavities.
    [Show full text]
  • The Effects of Learning Rate Schedules on Color Quantization
    THE EFFECTS OF LEARNING RATE SCHEDULES ON COLOR QUANTIZATION by Garrett Brown A thesis presented to the Department of Computer Science and the Graduate School of the University of Central Arkansas in partial fulfillment of the requirements for the degree of Master of Science in Computer Science Conway, Arkansas May 2020 ProQuest Number:27834147 All rights reserved INFORMATION TO ALL USERS The quality of this reproduction is dependent on the quality of the copy submitted. In the unlikely event that the author did not send a complete manuscript and there are missing pages, these will be noted. Also, if material had to be removed, a note will indicate the deletion. ProQuest 27834147 Published by ProQuest LLC (2020). Copyright of the Dissertation is held by the Author. All Rights Reserved. This work is protected against unauthorized copying under Title 17, United States Code Microform Edition © ProQuest LLC. ProQuest LLC 789 East Eisenhower Parkway P.O. Box 1346 Ann Arbor, MI 48106 - 1346 TO THE OFFICE OF GRADUATE STUDIES: The members of the Committee approve the thesis of Garrett Brown presented on March 31, 2020. M. Emre Celebi, Ph.D., Committee Chairperson Ahmad Patooghy, Ph.D. Mahmut Karakaya, Ph.D. PERMISSION Title The Effects of Learning Rate Schedules on Color Quantization Department Computer Science Degree Master of Science In presenting this thesis/dissertation in partial fulfillment of the requirements for a graduate degree from the University of Central Arkansas, I agree that the Library of this University shall make it freely available for inspections. I further agree that permission for extensive copying for scholarly purposes may be granted by the professor who supervised my thesis/dissertation work, or, in the professor’s absence, by the Chair of the Department or the Dean of the Graduate School.
    [Show full text]
  • A New Steganographic Method for Palette-Based Images
    IS&T's 1999 PICS Conference A New Steganographic Method for Palette-Based Images Jiri Fridrich Center for Intelligent Systems, SUNY Binghamton, Binghamton, New York Abstract messages. The gaps in human visual and audio systems can be used for information hiding. In the case of images, the In this paper, we present a new steganographic technique human eye is relatively insensitive to high frequencies. This for embedding messages in palette-based images, such as fact has been utilized in many steganographic algorithms, GIF files. The new technique embeds one message bit into which modify the least significant bits of gray levels in one pixel (its pointer to the palette). The pixels for message digital images or digital sound tracks. Additional bits of embedding are chosen randomly using a pseudo-random information can also be inserted into coefficients of image number generator seeded with a secret key. For each pixel transforms, such as discrete cosine transform, Fourier at which one message bit is to be embedded, the palette is transform, etc. Transform techniques are typically more searched for closest colors. The closest color with the same robust with respect to common image processing operations parity as the message bit is then used instead of the original and lossy compression. color. This has the advantage that both the overall change The steganographer’s job is to make the secretly hidden due to message embedding and the maximal change in information difficult to detect given the complete colors of pixels is smaller than in methods that perturb the knowledge of the algorithm used to embed the information least significant bit of indices to a luminance-sorted palette, except the secret embedding key.* This so called such as EZ Stego.1 Indeed, numerical experiments indicate Kerckhoff’s principle is the golden rule of cryptography and that the new technique introduces approximately four times is often accepted for steganography as well.
    [Show full text]
  • Good Colour Maps: How to Design Them
    Good Colour Maps: How to Design Them Peter Kovesi Centre for Exploration Targeting School of Earth and Environment The University of Western Australia Crawley, Western Australia, 6009 [email protected] September 2015 Abstract Many colour maps provided by vendors have highly uneven percep- tual contrast over their range. It is not uncommon for colour maps to have perceptual flat spots that can hide a feature as large as one tenth of the total data range. Colour maps may also have perceptual discon- tinuities that induce the appearance of false features. Previous work in the design of perceptually uniform colour maps has mostly failed to recognise that CIELAB space is only designed to be perceptually uniform at very low spatial frequencies. The most important factor in designing a colour map is to ensure that the magnitude of the incre- mental change in perceptual lightness of the colours is uniform. The specific requirements for linear, diverging, rainbow and cyclic colour maps are developed in detail. To support this work two test images for evaluating colour maps are presented. The use of colour maps in combination with relief shading is considered and the conditions under which colour can enhance or disrupt relief shading are identified. Fi- nally, a set of new basis colours for the construction of ternary images arXiv:1509.03700v1 [cs.GR] 12 Sep 2015 are presented. Unlike the RGB primaries these basis colours produce images whereby the salience of structures are consistent irrespective of the assignment of basis colours to data channels. 1 Introduction A colour map can be thought of as a line or curve drawn through a three dimensional colour space.
    [Show full text]
  • Gradient Magnitude Similarity Deviation: an Highly Efficient Perceptual Image Quality Index
    1 Gradient Magnitude Similarity Deviation: An Highly Efficient Perceptual Image Quality Index Wufeng Xue, Lei Zhang, Member IEEE, Xuanqin Mou, Member IEEE, and Alan C. Bovik, Fellow, IEEE Abstract—It is an important task to faithfully evaluate the perceptual quality of output images in many applications such as image compression, image restoration and multimedia streaming. A good image quality assessment (IQA) model should not only deliver high quality prediction accuracy but also be computationally efficient. The efficiency of IQA metrics is becoming particularly important due to the increasing proliferation of high-volume visual data in high-speed networks. Figure 1 The flowchart of a class of two-step FR-IQA models. We present a new effective and efficient IQA model, called ratio (PSNR) correlates poorly with human perception, and gradient magnitude similarity deviation (GMSD). The image gradients are sensitive to image distortions, while different local hence researchers have been devoting much effort in structures in a distorted image suffer different degrees of developing advanced perception-driven image quality degradations. This motivates us to explore the use of global assessment (IQA) models [2, 25]. IQA models can be classified variation of gradient based local quality map for overall image [3] into full reference (FR) ones, where the pristine reference quality prediction. We find that the pixel-wise gradient magnitude image is available, no reference ones, where the reference similarity (GMS) between the reference and distorted images image is not available, and reduced reference ones, where combined with a novel pooling strategy – the standard deviation of the GMS map – can predict accurately perceptual image partial information of the reference image is available.
    [Show full text]
  • Fast and Optimal Laplacian Solver for Gradient-Domain Image Editing Using Green Function Convolution
    Fast and Optimal Laplacian Solver for Gradient-Domain Image Editing using Green Function Convolution Dominique Beaini, Sofiane Achiche, Fabrice Nonez, Olivier Brochu Dufour, Cédric Leblond-Ménard, Mahdis Asaadi, Maxime Raison Abstract In computer vision, the gradient and Laplacian of an image are used in different applications, such as edge detection, feature extraction, and seamless image cloning. Computing the gradient of an image is straightforward since numerical derivatives are available in most computer vision toolboxes. However, the reverse problem is more difficult, since computing an image from its gradient requires to solve the Laplacian equation (also called Poisson equation). Current discrete methods are either slow or require heavy parallel computing. The objective of this paper is to present a novel fast and robust method of solving the image gradient or Laplacian with minimal error, which can be used for gradient-domain editing. By using a single convolution based on a numerical Green’s function, the whole process is faster and straightforward to implement with different computer vision libraries. It can also be optimized on a GPU using fast Fourier transforms and can easily be generalized for an n-dimension image. The tests show that, for images of resolution 801x1200, the proposed GFC can solve 100 Laplacian in parallel in around 1.0 milliseconds (ms). This is orders of magnitude faster than our nearest competitor which requires 294ms for a single image. Furthermore, we prove mathematically and demonstrate empirically that the proposed method is the least-error solver for gradient domain editing. The developed method is also validated with examples of Poisson blending, gradient removal, and the proposed gradient domain merging (GDM).
    [Show full text]
  • Certified Digital Designer Professional Certification Examination Review
    Digital Imaging & Editing and Digital & General Photography Certified Digital Designer Professional Certification Examination Review Within this presentation – We will use specific names and terminologies. These will be related to specific products, software, brands and trade names. ADDA does not endorse any specific software or manufacturer. It is the sole decision of the individual to choose and purchase based on their personal preference and financial capabilities. the Examination Examination Contain at Total 325 Questions 200 Questions in Digital Image Creation and Editing Image Editing is applicable to all Areas related to Digital Graphics 125 Question in Photography Knowledge and History Photography is applicable to General Principles of Photography Does not cover Photography as a General Arts Program Examination is based on entry level intermediate employment knowledge Certain Processes may be omitted that are required to achieve an end result ADDA Professional Certification Series – Digital Imaging & Editing the Examination Knowledge of Graphic and Photography Acronyms Knowledge of Graphic Program Tool Symbols Some Knowledge of Photography Lighting Ability to do some basic Geometric Calculations Basic Knowledge of Graphic History & Theory Basic Knowledge of Digital & Standard Film Cameras Basic Knowledge of Camera Lens and Operation General Knowledge of Computer Operation Some Common Sense ADDA Professional Certification Series – Digital Imaging & Editing This is the Comprehensive Digital Imaging & Editing Certified Digital Designer Professional Certification Examination Review Within this presentation – We will use specific names and terminologies. These will be related to specific products, software, brands and trade names. ADDA does not endorse any specific software or manufacturer. It is the sole decision of the individual to choose and purchase based on their personal preference and financial capabilities.
    [Show full text]
  • A Framework for Detection of Linear Gradient Filled Regions and Their
    21st International Conference on Computer Graphics, Visualization and Computer Vision 2013 A framework for detection of linear gradient filled regions and their reconstruction for vector graphics Ruchin Kansal Prof. Subodh Kumar Department of computer Department of computer science science IIT Delhi IIT Delhi India, New Delhi India, New Delhi [email protected] [email protected] ABSTRACT Linear gradients are commonly applied in non-photographic art-work for shading and other artistic effects. It is sometimes necessary to generate a vector graphics form of raster images comprising such art-work with the expec- tation to obtain a simple output that is further editable, say, using any popular vector editing software. Further, this vectorization process should be automatic with minimal user intervention. We present a simple image vectorization algorithm that meets these requirements by detecting linear gradient filled regions and reconstructing the vector definition using that information. It uses a novel interval gradient optimization scheme to derive large regions of uniform gradient. We also demonstrate the technique on noisy images. Keywords Image vectorization Gradient reconstruction Region detection With the increasing demand for vector content, the need for converting raster images into vectors has also in- creased. This process is called Vectorization. We set out to obtain automatic vectorization with minimal human intervention even on potentially noisy images. Later re-rasterization of our vector representation should pro- duce the appearance of the original image at various (a) Original Image (b) Adobe Livetrace scales. Further, for wide-spread usage, this vector rep- output resentation should be in a standard form and be ed- itable.
    [Show full text]
  • Colour Image Gradient Regression Reintegration
    Colour Image Gradient Regression Reintegration Graham D. Finlayson1, Mark S. Drew2 and Yasaman Etesam2 1 School of Computing Sciences, University of East Anglia, Norwich, NR4 7TJ, U.K. [email protected] 2School of Computing Science, Simon Fraser University, Vancouver, British Columbia, Canada V5A 1S6, {mark,yetesam}@cs.sfu.ca Abstract much-used method for the reintegration task is the well known Suppose we process an image and alter the image gradients Frankot-Chellappa algorithm [3]. This method solves a Poisson in each colour channel R,G,B. Typically the two new x and y com- equation by minimizing difference from an integrable gradient ponent fields p,q will be only an approximation of a gradient and pair in the Fourier domain. Of course, many other algorithms hence will be nonintegrable. Thus one is faced with the prob- have been proposed, e.g. [4, 5, 6, 7, 8, 9]. lem of reintegrating the resulting pair back to image, rather than Note in the first place that any reintegration method leaves a derivative of image, values. This can be done in a variety of ways, constant of integration to be set – an offset added to the resulting usually involving some form of Poisson solver. Here, in the case image – since we start off with derivatives which would zero out of image sequences or video, we introduce a new method of rein- any such offset. Its value must be handled through a heuristic of tegration, based on regression from gradients of log-images. The some kind. strength of this idea is that not only are Poisson reintegration arti- But more fundamentally many reintegration methods will facts eliminated, but also we can carry out the regression applied generate aliasing of various kinds such as creases or halos in the to only thumbnail images.
    [Show full text]
  • SHORT PAPERS.Indd
    VSMM 2008 VSMM 2008 – ofthe14 Digital Heritage–Proceedings Digital Heritage This volume contains the Short Papers presented at VSMM 2008, the 14th th International Conference on Virtual Systems and Multimedia which took Proceedings of the 14 International place on the 20 to 25 October 2008 in Limassol, Cyprus. The conference title was “Digital Heritage: Our Hi-tech-STORY for the Future, Technologies Conference on Virtual Systems to Document, Preserve, Communicate and Prevent the Destruction of our Fragile Cultural Heritage”. and Multimedia The conference was jointly organized by CIPA, the International ICOMOS Committee on Heritage Documentation and the Cyprus Institute. It also Short Papers hosted the 38th CIPA Workshop dedicated on e-Documentation and Standardization in Cultural Heritage and the second Euro-Med Conference on IT in Cultural Heritage. Through the Cyprus Institute, VSMM 2008 received the support of the Government of Cyprus and the European Commission and it was held under the Patronage of H. E. the President of the Republic of Cyprus. th International Conference on Virtual Systems andMultimedia Virtual Systems on Conference International 20–25 October 2008 Limassol, Cyprus M. Ioannides, A. Addison, A. Georgopoulos, L. Kalisperis (Editors) VSMM 2008 Digital Heritage Proceedings of the 14th International Conference on Virtual Systems and Multimedia Short Papers 20–25 October 2008 Limassol, Cyprus M. Ioannides, A. Addison, A. Georgopoulos, L. Kalisperis (Editors) Marinos Ioannides Editor-in-Chief Elizabeth Jerem Managing Editor
    [Show full text]
  • Unmixing-Based Soft Color Segmentation for Image Manipulation
    Unmixing-Based Soft Color Segmentation for Image Manipulation YAGIZ˘ AKSOY ETH Zurich¨ and Disney Research Zurich¨ TUNC¸ OZAN AYDIN and ALJOSAˇ SMOLIC´ Disney Research Zurich¨ and MARC POLLEFEYS ETH Zurich¨ We present a new method for decomposing an image into a set of soft color Additional Key Words and Phrases: Soft segmentation, digital composit- segments that are analogous to color layers with alpha channels that have ing, image manipulation, layer decomposition, color manipulation, color been commonly utilized in modern image manipulation software. We show unmixing, green-screen keying that the resulting decomposition serves as an effective intermediate image ACM Reference Format: representation, which can be utilized for performing various, seemingly unrelated, image manipulation tasks. We identify a set of requirements that Yagız˘ Aksoy, Tunc¸ Ozan Aydin, Aljosaˇ Smolic,´ and Marc Pollefeys. 2017. soft color segmentation methods have to fulfill, and present an in-depth Unmixing-based soft color segmentation for image manipulation. ACM theoretical analysis of prior work. We propose an energy formulation for Trans. Graph. 36, 2, Article 19 (March 2017), 19 pages. producing compact layers of homogeneous colors and a color refinement DOI: http://dx.doi.org/10.1145/3002176 procedure, as well as a method for automatically estimating a statistical color model from an image. This results in a novel framework for automatic and high-quality soft color segmentation that is efficient, parallelizable, and 1. INTRODUCTION scalable. We show that our technique is superior in quality compared to previous methods through quantitative analysis as well as visually through The goal of soft color segmentation is to decompose an image into a an extensive set of examples.
    [Show full text]
  • Color Representation and Quantization
    Lecture 2 Color Representation and Quantization Yao Wang Polytechnic Institute of NYU, Brooklyn, NY 11201 With contributions from Zhu Liu, AT&T Labs Most figures from Gonzalez/Woods: Digital Image Processing Lecture Outline • Color perception and representation – Human ppperception of color – Trichromatic color mixing theory – Different color representations • ClColor image dildisplay – True color image – Indexed color images • Pseudo color images • Quantization Fundamentals – Uniform, non-uniform, dithering • Color quantization Yao Wang, NYU-Poly EL5123: color and quantization 2 Light is part of the EM wave Yao Wang, NYU-Poly EL5123: color and quantization 3 Illuminating and Reflecting Light • Illuminating sources (primary light): – emit light (e.g. the sun, light bulb, TV monitors) – perceived color depends on the emitted freq. – follow additive rule • R+G+B=White • Reflecting sources (secondary light): – reflect an incoming light (e.g. the color dye, matte surface, cloth) – perceived color depends on reflected freq (=emitted freq-absorbed freq.) – follow subtractive rule • R+G+B=Black Yao Wang, NYU-Poly EL5123: color and quantization 4 Eye vs. Camera Camera components Eye components Lens Lens, cornea Shutter Iris, pupil Film RtiRetina Cable to transfer images Optic nerve send the info to the brain Yao Wang, NYU-Poly EL5123: color and quantization 5 Human Perception of Color • Retina contains receptors – Cones • Day vision, can perceive color tone • Red, green, and blue cones – Rods • Night vision, perceive brightness only • Color sensation
    [Show full text]