LIBRARY OF CONGRESS
`
`Office of Business Enterprises
`Duplication Services Section
`
`THIS IS TO CERTIFY that the collections of the Library of Congress contain a bound
`volume entitled JPEG STILL IMAGE DATA COMPRESSION STANDARD, call number
`TA 1632.P45 1992, Copy 2. The attached — Cover Page, Title Page, Copyright Page, Table of
`Contents Pages, Chapter 2, Chapter 5 and Appendix A - are a true and complete representation
`from that work.
`
`THIS IS TO CERTIFY FURTHER, that work is marked with a Library of Congress
`Copyright Office stamp dated November 23, 1992.
`
`IN WITNESS WHEREOF, the seal of the Library of Congress is affixed hereto on
`May 18, 2018.
`
`Deirdre Scott
`Business Enterprises Officer
`Office of Business Enterprises
`Library of Congress
`
`101 Independence Avenue, SE Washington, DC 20540-4917 Tel 202.707.5650 www.loc.gov- duplicationservices@loc.gov
`
`HULU LLC
`Exhibit 1008
`IPR2018-01187
`Page 1
`
`Comcast - Exhibit 1008, page 1
`
`

`

`SS3119N100 AO NInnien
`'l1111
`1
`11
`1
`
`IIIIIIIII
`
`II
`
`(°I 2.6EilfiriSTO0D
`
`Z abed
`
`Comcast - Exhibit 1008, page 2
`
`

`

`JPEG
`
`STILL IMAGE
`DATA COMPRESSION
`STANDARD
`
`William B. Pennebaker
`Joan L. Mitchell
`
`VAN NOSTRAND REINHOLD
`New York
`
`Page 3
`
`Comcast - Exhibit 1008, page 3
`
`

`

`Copyright © 1993 by Van Nostrand Reinhold
`
`Library of Congress Catalog Card Number 92-32860
`ISBN 0-442-01272-I
`
`All rights reserved. No part of this work covered by
`the copyright hereon may be reproduced or used in any
`form by any means—graphic, electronic, or
`mechanical, including photocopying, recording, taping,
`or information storage and retrieval systems—without
`written permission of the publisher.
`
`Printed in the United States of America
`
`Van Nostrand Reinhold
`115 Fifth Avenue
`New York, Ncw York 10003
`
`Chapman and Hall
`2-6 Boundary Row
`I.ondon, SF.1 81-1N, England
`
`Thomas Nelson Australia
`102 Dodds Street
`South Melbourne 3205
`Victoria, Australia
`
`Nelson Canada
`1120 Birchmount Road
`Scarborough, Ontario M1K 5G4, Canada
`
`16 15 14 13 12 11 10 9 8 7 6 5 4 3 2 I
`
`Library of Congress Cataloging-in-Publication Data
`
`Pennebaker. William B.
`JPEG still image data compression standard / William B.
`Pennebaker. Joan L. Mitchell.
`p.
`em.
`Includes bibliographical references and index.
`ISBN 0-442-01272-1
`1 . Image processing—Digital techniques—Standards. 2. Data
`compression (Telecommunication)—Standards. 3. Algorithms.
`I. Mitchell, Joan L. II. Title.
`TA1632.P45 1992
`621.36'7—dc20
`
`92-32860
`CIP
`
`Page 4
`
`Comcast - Exhibit 1008, page 4
`
`

`

`CONTENTS
`
`FOREWORD
`
`xiii
`
`ACKNOWLEDGMENTS
`
`xv
`
`TRADEMARKS
`
`xvii
`
`1
`CHAPTER 1. INTRODUCTION
`1.1 Examples of JPEG image compression 0
`4
`1.2 Organization of the book 0
`1.3 An architecture for image compression 0
`1.4 JPEG baseline and extended systems 0
`7
`1.5 An evolving standard 0
`1.6 An international collaboration 0
`
`7
`
`4
`
`6
`
`6
`
`CHAPTER 2. IMAGE CONCEPTS AND VOCABULARY
`2.1 Digital images 0
`10
`2.2 Sampling 0
`10
`2.3 Two-dimensional arrays of samples 0
`13
`2.4 Digital image data types 0
`
`12
`
`9
`
`vii
`
`Page 5
`
`Comcast - Exhibit 1008, page 5
`
`

`

`viii
`
`JPEG: STILL IMAGE DATA COMPRESSION STANDARD
`
`2.5 Large amounts of data 0
`
`21
`
`CHAPTER 3. ASPECTS OF THE HUMAN VISUAL SYSTEM
`23
`3.1 Luminance sampling 0
`25
`3.2 Sample precision 0
`3.3 Chrominance sampling C)
`25
`3.4 Linearity 0
`
`25
`
`CHAPTER 4. THE DISCRETE COSINE TRANSFORM (DCT)
`4.1 Basic DCI' concepts 0
`29
`4.2 Mathematical definition of the FDCT and IDCT a
`41
`4.3 Fast DCTs 3
`
`39
`
`23
`
`29
`
`65
`65
`
`73
`
`CHAPTER 5. IMAGE COMPRESSION SYSTEMS
`5.1 Basic structure of image compression systems 0
`67
`5.2 Image compression models 0
`5.3 JPEG entropy encoder and entropy decoder structures 0
`5.4 Transcoding 0
`77
`5.5 JPEG lossless and lossy compression 0
`5.6 Sequential and progressive coding 0
`5.7 Hierarchical coding 0
`79
`5.8 Compression measures 0
`
`78
`79
`
`79
`
`CHAPTER 6. JPEG MODES OF OPERATION
`6. 1 Sequential DCT-based mode of operation 0
`6.2 Progressive DCT-based mode of operation 0
`6.3 Sequential lossless mode of operation 0
`92
`6.4 Hierarchical mode of operation 0
`93
`
`81
`81
`86
`
`CHAPTER 7. JPEG SYNTAX AND DATA ORGANIZATION
`7.1 Control procedures and compressed data structure 0
`97
`7.2 Interchange and abbreviated compressed data formats 0
`7.3 Image data ordering 0
`99
`7.4 Marker definitions 0
`105
`7.5 Frame header 0
`110
`7.6 Scan header 0
`113
`7.7 Limit on the number of data units in an MCU 0
`7.8 Marker segments for tables and parameters 0
`7.9 Hierarchical progression marker segments 0
`7.10 Examples of JPEG data streams 0
`122
`7.11 Backus-Naur Form •
`127
`7.12 JPEG BNF •
`130
`
`116
`117
`120
`
`99
`
`CHAPTER 8. ENTROPY CODING CONCEPTS
`8.1 Entropy and information 0
`135
`8.2 An example to illustrate entropy coding 0
`8.3 Variable-length code words 0
`137
`8.4 Statistical modeling 0
`143
`8.5 Adaptive coding 0
`147
`
`135
`
`137
`
`97
`
`Page 6
`
`Comcast - Exhibit 1008, page 6
`
`

`

`CONTENTS
`
`ix
`
`CHAPTER 9. JPEG BINARY ARITHMETIC CODING
`9.1 The QM-encoder 3
`151
`9.2 The QM-decoder 0
`162
`9.3 More about the QM-coder 0
`
`166
`
`149
`
`CHAPTER 10. JPEG CODING MODELS
`169
`10. 1 JPEG sequential DCT-based coding models 0
`10.2 Models for progressive DCT-based coding 0
`10.3 Coding model for lossless coding 0
`182
`10.4 Models for hierarchical coding 0
`185
`
`170
`173
`
`CHAPTER 11. JPEG HUFFMAN ENTROPY CODING
`189
`11.1 Statistical models for the Huffman DCT-based sequential
`mode 0
`190
`11.2 Statistical models for progressive DCT-based coding 0
`194
`11.3 Statistical models for lossless coding and hierarchical mode spatial
`corrections 0
`198
`11.4 Generation of Hufftnan tables 0
`
`198
`
`CHAPTER 12. ARITHMETIC CODING STATISTICAL
`MODELS
`203
`12.1 Overview of JPEG binary arithmetic-coding procedures 0
`204
`12.2 Decision trees and notation 0
`12,3 Statistical models for the DCT-based sequential mode with
`206
`arithmetic coding 0
`12.4 Statistical models for progressive DCT-based coding 0
`214
`12.5 Statistical models for lossless coding and hierarchical-mode spatial
`corrections 0
`216
`12.6 Arithmetic coding conditioning tables 0
`
`218
`
`203
`
`CHAPTER 13. MORE ON ARITHMETIC CODING
`13.1 Optimal procedures for hardware and software •
`13.2 Fast software encoder implementations •
`225
`13.3 Fast software decoder implementations •
`228
`13.4 Conditional exchange •
`229
`13.5 QM-coder versus Q-coder •
`229
`13,6 Resynchronization of decoders •
`231
`13.7 Speedup mode •
`
`230
`
`219
`220
`
`233
`
`CHAPTER 14. PROBABILITY ESTIMATION
`234
`14.1 Bayesian estimation 0
`235
`14.2 Renormalization-driven estimation 0
`14.3 Markov-chain modelling of the probability estimation •
`14.4 Approximate model •
`238
`14.5 Single- and mixed-context models •
`14.6 Single-context model •
`241
`14.7 Mixed-context model •
`243
`14.8 Application of the estimation models to the QM-coder CI
`247
`14.9 Initial learning 0
`
`240
`
`236
`
`244
`
`Page 7
`
`Comcast - Exhibit 1008, page 7
`
`

`

`JPEG: STILL IMAGE DATA COMPRESSION STANDARD
`
`14.10 Robustness of estimators versus refinement of models C11
`249
`14.1 1 Other estimation tables 01
`
`249
`
`CHAPTER 15. COMPRESSION PERFORMANCE
`254
`15.1 Results for baseline sequential DCT 0
`15.2 Results for sequential DCT with arithmetic coding 3
`15.3 Results for sequential DCT with restart capability 4
`15.4 Results for progressive DCT with arithmetic coding 3
`15.5 Results for lossless mode with arithmetic coding 0
`258
`15.6 Summary of Results 0
`
`253
`
`255
`255
`256
`257
`
`261
`CHAPTER 16. JPEG ENHANCEMENTS
`16.1 Removing blocking artifacts with AC prediction 0
`264
`16.2 Low hitrate VQ enhanced decoding 3
`16.3 An approximate form of adaptive quantization C)
`266
`16.4 Display-adjusted decoding 0
`
`261
`
`264
`
`CHAPTER 17. JPEG APPLICATIONS AND VENDORS
`269
`17.1 Adobe Systems Incorporated 0
`270
`17.2 AT&T Microelectronics 0
`17.3 AutoGraph International ApS 0
`272
`17.4 AutoView 0
`272
`17.5 Bulletin board systems 0
`17.6 California Department of Motor Vehicles 0
`17.7 C-Cube Microsystems, Inc. 0
`273
`17.8 Data Link 0
`275
`17.9 Discovery Technologies, Inc. 0
`17.10 DSP 0
`276
`17.11 Eastman Kodak Company 0
`17.12 Handmade Software 0
`277
`17.13 IBM 0
`278
`17.14 Identix 0
`279
`17.15 in o
`279
`17.16 Independent JPEG Group 0
`17.17 ITR 0
`281
`17.18 Lewis Siwell, Inc. 0
`17.19 LSI Logic 0
`282
`17.20 Moore Data Management Services 0
`17.21 NBS Imaging C)
`283
`17,22 NTT Electronics Technology Ltd. 0
`17.23 OPTI BASE® 0
`284
`17.24 Optivision, Inc. 0
`284
`17.25 Philips Kommunikations Industrie 0
`17.26 PRISM 0
`286
`17.27 Storm Technology 0
`286
`17.28 Telephoto Communications 0
`17.29 Tribune Publishing Co. 0
`288
`17.30 VideoTelecom 0
`288
`17.31 XImage 0
`289
`
`281
`
`283
`
`284
`
`285
`
`271
`
`273
`
`275
`
`276
`
`280
`
`287
`
`267
`
`Page 8
`
`Comcast - Exhibit 1008, page 8
`
`

`

`CONTENTS
`
`xi
`
`289
`17.32 Xing 0
`17.33 Zoran Corporation 0
`291
`17.34 3M 0
`17.35 File formats 0
`
`292
`
`290
`
`CHAPTER 18. OVERVIEW OF CCITT, ISO, AND IEC
`296
`18.1 ISO 0
`297
`18.2 CCITT 0
`298
`18.3 IEC 0
`18.4 Joint coordination 0
`
`299
`
`295
`
`CHAPTER 19. HISTORY OF JPEG
`19.1 Formation of JPEG 0
`301
`19.2 Original JPEG Goals 0
`302
`19.3 Selecting an approach 0
`302
`19.4 Functional requirements 0
`303
`19.5 Refining the ADCT technique 0
`19.6 Technical specifications 0
`307
`19.7 ISO 10918 Part 1 0
`309
`19.8 JPEG Part 1 DIS ballot results 0
`19.9 CCITT Recommendation T.81 0
`311
`19.10 ISO 10918 Part 2 0
`19.11 JPEG Goals Achieved 0
`
`313
`
`301
`
`306
`
`311
`311
`
`CHAPTER 20. OTHER IMAGE COMPRESSION STANDARDS
`317
`20.1 CCITT G3 and G4 0
`318
`20.2 H.261 0
`20.3 JBIG 0
`318
`20.4 MPEG 0
`325
`
`317
`
`CHAPTER 21. POSSIBLE FUTURE JPEG DIRECTIONS
`21.1 Adaptive quantization 0
`331
`21.2 Improvements to lossless coding t
`333
`21.3 Other possible addenda 0
`21.4 Backwards compatibility 0
`333
`
`332
`
`331
`
`APPENDIX A. ISO DIS 10918-1 REQUIREMENTS AND
`GUIDELINES
`335
`
`APPENDIX B. DRAFT ISO DIS 10918-2 COMPLIANCE
`TESTING
`545
`
`REFERENCES
`
`627
`
`INDEX
`
`632
`
`Page 9
`
`Comcast - Exhibit 1008, page 9
`
`

`

`2
`
`IMAGE CONCEPTS
`AND VOCABULARY
`
`In this chapter we develop some of the basic concepts and vocabulary
`needed for an understanding of the image compression functions contained
`in JPEG. Our approach in this chapter will be qualitative rather than
`mathematical.
`The reader undoubtedly has an intuitive grasp of what the term
`"image" means. We are surrounded by images: photographs, television,
`printed material, etc., all aimed at the incredibly powerful visual input
`channel into the human brain. We arc used to thinking of images as two-
`dimensional, but in fact, they often have several more dimensions. Color,
`time (motion), and depth (three-dimensional and stereo pairs) can add
`additional dimensions to the information in an image, and, if we want to
`consider the image as a fractal, the question of dimensionality becomes yet
`more complex. By definition, however, JPEG is concerned primarily with
`images that have two spatial dimensions, contain grayscalc or color infor-
`mation, and possess no temporal dependence (hence the word "still" in the
`title of the JPEG standard).
`
`9
`
`Page 10
`
`Comcast - Exhibit 1008, page 10
`
`

`

`10
`
`JPEG: STILL IMAGE DATA COMPRESSION STANDARD
`
`2.1 Digital images 0
`The images compressed with the JPEG techniques are digital images,
`Many readers are already familiar with the way in which television images
`are scanned. Starting at the top of the screen, a television field is displayed
`on a CRT (cathode-ray tube) by the scanning of an electron beam one
`horizontal line at a time in rapid sequence. An image created in this way
`is already segmented into a set of discrete scan lineS. To turn this into a
`true digital image we do two more things! take samples along each scan line
`at regular intervals and convert these samples into binary numbers which
`can he stored in a computer.
`
`2.2 Sampling 0
`The sampling process used to create a digital image necessarily discards
`an enormous amount of information: the fine spatial detail and tiny
`amplitude variations that are, by definition, not important enough to pre-
`serve. The term lossIess compression, which we shall discuss later in this
`chapter implies preservation of all information, but in fact such com-
`pression preserves only the information in the sampled data, totally
`ignoring the loss of information that is a necessary part of capturing any
`digital image.
`There are three important parameters in sampling a signal: precision,
`sampling interval, and sampling aperture.
`Suppose we have the one-dimensional analog signal, perhaps the out-
`put of a television camera, shown in Figure 2-1a. When we take meas-
`urements of this signal at discrete points, marked by dots, we are sampling
`the signal. When we convert to a digital representation of the signal we
`are expressing the sample with some fixed number of bits per sample,
`which is the precision of the sample. In the process we are assuming that
`this fixed number of bits is sufficient to represent the signal, much as we
`restrict the number of decimal places when we write down a number, The
`only difference is that we use binary digits or "bits," the traditional way
`of expressing numbers in computers.
`The interval between samples determines the resolution of the sam-
`pled signal. For the signal in Figure 2-la resolution can be defined as the
`number of samples taken per unit time. For a given precision in our
`samples, the higher the resolution (that is, the more samples we take in a
`given time interval), the more accurately we can represent the variations
`in the signal. The resolution thus provides an upper limit on the fre-
`quencies that can be represented in the sampled signal. Just how we sam-
`ple a signal is very important, however, because we can get very misleading
`measurements if we do it improperly.
`Figure 2-lb illustrates a signal that varies rapidly between samples.
`If we connect the dots as shown in Figure 2-lb, the sampled output appears
`to have a much lower frequency than the original analog signal. This dis-
`tortion is termed aliasing, and is closely related to an effect seen in
`western -movie scenes, in which wagon wheels appear to turn backwards.
`Figure 2-lb illustrates an important point. You must have enough samples
`
`Page 11
`
`Comcast - Exhibit 1008, page 11
`
`

`

`CHAPTER 2. IMAGE CONCEPTS AND VOCABULARY
`
`11
`
`(a)
`
`(b)
`
`(c)
`
`Figure 2-1. Sampling illustrations. (a) Illustrates sampling of a one-
`dimensional analog signal. (b) Illustrates how aliasing can
`occur when a high-frequency signal is sampled too infre-
`quently. (c) Shows the suppression of aliasing when the sig-
`nal is averaged over a window equal to the sampling interval.
`
`to represent properly a signal that is changing., When you do not have
`enough samples, you may get aliasing effects.
`Aliasing can he largely suppressed if the input signal is averaged over
`an interval (the aperture) that is roughly equal to the interval between
`
`Page 12
`
`Comcast - Exhibit 1008, page 12
`
`

`

`JPEG: STILL IMAGE DATA COMPRESSION STANDARD
`
`12
`samples. Figure 2-1c illustrates this, Optical scanning input devices often
`do this averaging automatically in the process of scanning an image, usu-
`ally because the effective aperture (opening) through which the light is
`acquired by each sensor is nearly as wide as the spatial interval between
`light sensors. However, even the best image input devices will exhibit
`minor aliasing artifacts.
`One .of the common space- and cost-saving techniques, 'especially in
`working with images, is to reduce the spatial resolution of the data
`the number of samples taken in a given space). This can be done in a
`number of ways, including a simple method known as "subsampling." In
`subsampling by a factor of two, for example, every other sample is dis-
`carded, thereby halving the number of samples representing the data. Note
`that this effectively makes the sampling interval twice as large, while
`leaving the aperture unchanged. The net result is equivalent to sampling
`with too small an aperture. Sampling with too small an aperture or too
`large a sampling interval, as was done in Figure 2-lb, will provide an
`insufficient number of samples for a truly accurate representation of the
`original image, and can result in severe aliasing effects.
`The solution to this problem is to use an average of two or more
`samples rather than a single sample in the process of reducing the resol-
`ution. We call this process filtering, and the weight we give to each sample
`as we average them is called the filter coefficient. The only restriction is
`that these weights should sum to one. Figure 2-2 gives two simple weighted
`averages (filters) that might be used when subsampling by a factor of two
`in one dimension. Note that the center of each subsampled point labelled
`Y coincides with the center of one of the original sampling points labelled
`X; the center of each subsampled point labelled Z coincides with a
`boundary between sample points labelled X. These are examples of two
`different sample registrations.
`Taking a weighted average is also known as low-pass ‘filtering, because
`the effect of the averaging is to remove high spatial frequencies. These
`filters are also called downsampling filters. With different weights one can
`also high-pass filter, i.e., selectively enhance the high spatial frequencies.
`High-pass filters are also known as upsampling filters.
`
`2.3 Two-dimensional arrays of samples 0
`
`For simplicity let us consider a digital image obtained by scanning a
`monochrome (grayscale) photograph. We create a two-dimensional array
`of samples, each sample with 8-bit precision. Although sampling grids
`other than rectangular are possible, MEG has chosen to use only simple
`rectangular grids; we shall therefore limit our discussion to such grids.
`We shall have more to say about precision requirements in following
`sections, but one sample per byte (eight bits) is a convenient precision for
`computer storage of images. This precision is usually sufficient to give the
`illusion of a continuous range of grayscale values when the image is
`observed on a typical CRT display. Some applications, notably those
`dealing with medical images, may require higher precision (12 bits/sample
`or more).
`
`Page 13
`
`Comcast - Exhibit 1008, page 13
`
`

`

`CHAPTER 2. IMAGE CONCEPTS AND VOCABULARY
`
`13
`
`MOM
`
`Source
`samples
`
`-0-- Weights
`
`Down-sampled
`samples
`
`(a)
`
`(b)
`
`Figure 2-2. Two examples of low-pass filtering
`
`Each sample in a monochrome image is called a pixel or pel. Pixel
`is a contraction of "picture element" and is used in the display industry.
`Pd is a contraction for "print element" and is used in the printing industry.
`In modern usage the two are almost interchangeable, and the distinction
`between them has been lost. This is probably fortunate, since in many
`applications there is a need both to display and to print images. Note that
`in color images a set of samples is usually required to represent a single
`
`Precision and aspect ratio are two attributes of a sample. Precision
`determines how many levels of intensity can be represented and is
`expressed as the number of bits/sample. Aspect ratio describes the shape
`of the sample, and this is determined by the relationship between the
`physical size of the image and the rectangular grid of samples. If the shape
`of the sample is not the same between input and output devices, geometric
`distortion will result.
`Figure 2-3 illustrates the appearance of an image at three different
`precisions. Figure 2-3a shows an image at 8 bits/sample, Figure 2-3b shows
`the same image at one bit/sample and Figure 2-3c shows this image at four
`bits/sample.
`
`2.4 Digital image data types 0
`The examples in Figure 2-3 illustrate some of the different digital image
`data types. These are all examples of monochrome still images, where the
`
`' If an appropriate color-encoding format is used for the analog signal, single
`samples can be used to represent color information. However, JPEG does not
`apply to color images sampled in this format.
`
`Page 14
`
`Comcast - Exhibit 1008, page 14
`
`

`

`14
`
`JPEG: STILL IMAGE DATA COMPRESSION STANDARD
`
`(a)
`
`(b)
`
`Hotel Lotus-Lila
`Playa Laguna
`
`Figure 2-3. Examples of still-image precisions. (a) Eight bits/sample. (b)
`One bit/sample. (a) is the luminance component of the JPEG
`test image, "hotel".
`
`Page 15
`
`Comcast - Exhibit 1008, page 15
`
`

`

`CHAPTER 2. IMAGE CONCEPTS AND VOCABULARY
`
`15
`
`(0)
`
`Figure 2-3 continued. (c) Four bits/sample.
`
`term "still" is used to distinguish from the set of images that make up a
`motion sequence. Unless otherwise stated, the images should be assumed
`to be "still" in this book.
`
`2.4.1 Grayscale 0
`At eight bits/sample, the example shown in Figure 2-3a has sufficient pre-
`cision to be considered a grayscale or continuous-tone image. JPEG is
`designed for this class of data—data that are of high enough precision to
`give the appearance of continuity to an observer. JPEG can be used to
`compress limited-precision data, but may not be as efficient as alternative
`approaches designed for such data. A discontinuous representation such
`as a color-palette image must be remapped to a continuous representation
`before JPEG can be used effectively.
`
`2.4.2 Binary image 0
`
`Figure 2-3b shows a binary or bi-level image that could be transmitted by
`a facsimile machine. These images have only one bit per sample and are
`usually images of text or line drawings in which the original has only two
`tones, black and white. The resolution of these images is usually very high,
`such that the eye would have trouble discerning subtle changes in grayscale
`value from sample to sample. At these higher resolutions, a process called
`digital halftoning creates the effect of continuous tones in a binary image
`by alternating between closely spaced black and white samples. The effect
`is similar to the technique used to reproduce photographs in newspapers.
`
`Page 16
`
`Comcast - Exhibit 1008, page 16
`
`

`

`16
`
`JPEG: STILL IMAGE DATA COMPRESSION STANDARD
`
`Indeed, photographs can be regarded as the most limiting form of
`halftoning, in that the photographic grains are either white or black.
`However, the effective resolution required to resolve single grains is on the
`order of 5,000 pels/inch (2,000 pels/cm). Therefore, even at the highest
`resolutions currently used for binary facsimile images (400 pels/inch),
`digital halftoning is an imperfect substitute for true grayscale capability.
`in Chapter 20 is
`The JBIG Draft International Standard discussed
`designed for these binary images.
`
`2.4.3 Limited bits/sample 0
`Figure 2-3c is an example of a lower-precision or limited bits/sample for-
`mat that is typical of computer graphics. The bound between limited
`hits/sample and continuous tone is not rigorously defined, but can be taken
`to be about four bits/sample.
`At intermediate resolutions, images of black/white documents can
`benefit from additional precision. A limited number of bits per sample can
`provide for grayscale in transition regions between black and white. This
`greatly improves the apparent quality of the images. The use of grayscale
`in this manner is sometimes called anti-aliasing and can be an effective tool
`for removing the "jaggies" in graphics images.
`
`2.4.4 Color 0
`
`According to the trichromatic theory, the sensation of color is produced
`by selectively exciting three classes of receptors in the eye. Certain fre-
`quencies in the visible light spectrum will excite certain receptors,
`producing the effect of color. Furthermore, if we provide the same stim-
`ulus to these receptors from two different sources, the two sources will
`appear to have the same color.
`In color imaging there are two basic ways of producing this selective
`excitation: additive color and subtractive color. Additive color is used
`with active light-emitting systems in which the light from sources of dif-
`ferent colors is added together to produce the perceived color. Subtractive
`color is used with passive systems, in which light from a given source is
`selectively absorbed at different wavelengths, leaving only the wavelengths
`that will be perceived as the desired colors. Color can also be produced
`by other physical processes, such as refraction and interference, but these
`are simply alternative ways of creating additive or subtractive color.
`In an additive color device such as a display CRT, the light is
`produced by three primary phosphors, red, green, and blue (ROB). These
`phosphors are excited separately by the electron beam in the CRT, and the
`light emitted from the three phosphors stimulates the three types of
`receptors in the eye to produce the perception of color. When excited in
`the right proportion, the three phosphors produce the perception of white
`light; the absence of excitation produces black.
`Subtractive color is used in the printing industry. Three colors (and
`sometimes a fourth) are superimposed on a white reflective surface such
`as paper. The inks, typically cyan (a blue-green), magenta, and yellow
`(CMY), selectively absorb certain ranges of wavelengths of light. The eye
`
`Page 17
`
`Comcast - Exhibit 1008, page 17
`
`

`

`CHAPTER 2. IMAGE CONCEPTS AND VOCABULARY
`
`17
`
`Line where
`R=G=B (Gray values)
`
`Black
`
`G
`
`B
`
`Figure 2-4. RGB (red-green-blue) color coordinate system. On the line
`labelled R=G=B are shades of gray ranging from black to
`white.
`
`perceives the reflected light, which has not been absorbed; hence the term
`"subtractive." Where no ink is on the paper the light reflected is white;
`where all three inks are present, the light is (in principle) absorbed and the
`appearance is black. In practice, complete absorption is difficult to achieve
`in printing inks and a fourth ink, black (thus, CMYK, in which K stands
`for blacK), is used as well.
`
`2.4.4.1 Color spaces/coordinates 0
`Many representations of color images are possible. The trichromatic the-
`ory tells us that, ideally, three arrays of samples (three components) should
`he sufficient to represent a color image.2 However, output devices are
`limited, so not all colors can be obtained.
`RGB is one example of a color representation requiring three inde-
`pendent values to describe the colors. Each of the values can be varied
`independently, and we can therefore create a three-dimensional space with
`the three components, R, G, and B, as independent coordinates. Colors
`are represented as points in this space, as shown in Figure 2-4. Note that
`shades of gray from black to white are found on the diagonal line in this
`plot. In general pixels in a color image have information from the samples
`of each component, and the color image is comprised of the two-
`dimensional arrays of the component samples.
`Color representations such as RGB or CMY are not always the most
`convenient. Other representations are available that use color components
`
`Page 18
`
`Comcast - Exhibit 1008, page 18
`
`

`

`18
`
`JPEG: STILL IMAGE DATA COMPRESSION STANDARD
`
`that are closely related to the criteria used to describe color perception:
`brightness, hue, and saturation. Brightness describes the intensity of the
`light (revealing whether it is white, gray. or black) and this can be related
`to the luminance of the source. Hue describes what color is present (red,
`green, yellow, etc.) and this can be related to the dominant wavelength of
`the light source. Saturation describes how vivid the color is (very strong,
`pastel, nearly white) and this can he related to the purity or narrowness
`of the spectral distribution of the source.
`Color spaces or color coordinate systems in which one component is
`the luminance and the other two components are related to hue and satu-
`ration arc called luminance-chrominance representations. The luminance
`provides a grayscale version of the image (such as the image on a
`monochrome television receiver), and the chrominance components pro-
`vide the extra information that converts the grayscale image to a color
`image. Luminance-chrominance representations are particularly important
`for good image compression.
`
`2.4.4.2 Linear color transformations 0
`The luminance of a display system is the sum of the luminances of the red,
`green, and blue phosphors. For convenience, the values of the three colors,
`R, G, and B, are expressed by a relative scale from 0 to 1, where 0 indicates
`no excitation of the phosphor and 1 indicates maximum excitation. In
`addition, the display is adjusted so that a gray value between black and
`white results whenever the three primary signals, R (red), G (green) and B
`(blue), are equal. For a particular definition of red, green and blue, the
`luminance (Y) of any color can be calculated from the following weighted
`sum:3
`
`Y= 0.3R + 0.6G +0.1B
`
`[2-1]
`
`The scaling is chosen such that the luminance is also expressed by a relative
`scale from 0 to 1 and the weights reflect the contributions of the individual
`primaries to the total luminance.
`The term chrominance is defined as the difference between a color and
`a reference white at the same luminance. The chrominance information
`can therefore be expressed by a set of color differences, V and U, where V
`and U are defined by:
`
`R — Y
`
`U=B — Y
`
`[2-2]
`
`[2-3]
`
`These color differences are zero whenever R=G--B, as this condition
`produces gray. which has no chrominance. The V component controls
`colors ranging from red (V> 0) to blue-green (V< 0), whereas the U com-
`ponent controls colors ranging from blue (Li > 0)
`to yellow (U < 0).
`Together with the luminance, these chrominance coordinates make up the
`color coordinate system known as YUV. Figure 2-5 illustrates the
`relationship between this YUV color space and the RGB color space.
`
`Page 19
`
`Comcast - Exhibit 1008, page 19
`
`

`

`CHAPTER 2. IMAGE CONCEPTS AND VOCABULARY
`
`19
`
`Plane for Y=0.3
`
`Plane for U=0
`
`Y=1
`
`Plane for V=0
`
`Figure 2-5. Relationship between the RGB and YUV coordinate systems
`
`Chrominance values of zero are located on the diagonal where R= G = B,
`and the realizable range of YUV values is bounded by the RGB cube. The
`YUV color space is used in the European television systems.
`Another color space, YIQ, is used in the North American television
`systems. The YIQ space is related to the YUV space as follows: l' is the
`same for both spaces whereas I and Q are related only to U and V. I and
`Q are given by:
`
`1=0.74V— 0.27U
`
`Q=0.48V+0.41U
`
`[2-4]
`
`[2-5]
`
`Still another color coordinate system, YCbCr, was used extensively in
`the development of the JPEG standard. This color coordinate system is
`closely related to YUV. It uses the same Y. coordinate as the YUV system,
`whereas U and V are scaled and zero-shifted to produce the variables Cb
`and Cr, respectively. The equations are:
`
`Ch = (U12) + 0.5
`
`Cr=(V/I.6)+ 0.5
`
`[2-6]
`
`[2-7]
`
`Page 20
`
`Comcast - Exhibit 1008, page 20
`
`

`

`20
`
`JPEG: STILL IMAGE DATA COMPRESSION STANDARD
`
`With this scaling and zero shifting the chrominance values arc always in
`the range 0 to I. The values of Cb and Cr are sometimes multiplied by
`255, so that they can be represented by an 8-hit integer.
`In principle, a number of other color spaces, including the CIE
`(Commission Internationale de l'Eclairage) tristimulus values, can he
`related to these spaces by simple linear equations.
`2.4.4.3 Uniform perceptual color spaces 0
`Although luminance-chrominance coordinates such as YUV are very con-
`venient from the standpoint of ease of conversion to and from RGB, they
`suffer from one drawback: the amount of perceived color change produced
`by a fixed small change in these coordinates is quite nonuniform, and
`depends on the particular values of luminance and chrominance. Two
`color spaces, CIELIJV and CIELAB, which provide a relatively uniform
`perceptual space for describing colors, have been defined. A uniform per-
`ceived change for given input change is a very desirable property, in that
`the precision required to express colors to a particular degree of fidelity
`can be more readily specified in a space with uniform perceptual charac-
`teristics. Unfortunately, the transformations between the linear spaces
`described in section 2.4.4.2 and these uniform perceptual spaces are rela-
`tively complex. Nonetheless, these spaces are expected to be used exten-
`sively in applications requiring precise color reproduction. Accurate color
`reproduction is a very complex subject and a complete discussion of it is
`well beyond the scope of this book.
`
`2.4.4.4 Colorblindness of JPEG 0
`From the perspective of JPEG, the question of which coordinate sy