Errata list for Data Compression: The Complete Reference, 2nd Edition

Last Updated 8 May 2004. Errors found by the author. Page 126, following section 2.16 there should be a bibliography with: Pennebaker, W.~B., and J.~L. Mitchell (1988) ``Probability Estimation for the Q-coder,'' {\sl IBM Journal of Research and Development} {\bf32}(6):717--726. Pennebaker, W.~B., J.~L. Mitchell, et al. (1988) ``An Overview of the Basic Principles of the Q-coder Adaptive Binary Arithmetic Coder,'' {\sl IBM Journal of Research and Development} {\bf32}(6):737--752. The entire issue 6, 1988 of the IBM Journal of Research and Development was devoted to the Q-coder. Page 179, start of Section 3.10.1: "undrestand" should be "understand". Page 229, line +1: "compresseses" should be "compresses". Page 242, line -2 "image is thus 1/2" should be "image is thus 1/3" Page 242, line -1 "or 2/3 of the original size" should be "or 1/2 of the original size" Page 247, Line +2: 0.0012 should be 0.012. Page 247, line +6: "that forms 45^\circ angles" should be "that forms equal (although not 45^\circ) angles" Page 249, 2nd display equation, 2nd part: C'= W.D = 1/2 \matrix{1 1 ...}, D = ... should be   C'= W.D = 1/2 \matrix{1 1 ...}.D = ... " (This improvement to the original error correction was proposed by Horacio Luis Varela Page 249, the line above the last display equation: C=C'.W^T=W.C.W^T should be C=C'.W^T=W.D.W^T=W.D.W Page 250, line +10: "For of these reasons..." should be "For these reasons..." Page 266, line -11: ", but for of the reasons..." should be ", but for the reasons..." Page 476, line -4: "16=2(8-1)" should be "14=2(8-1)" Page 479, line -6: "columns of the image" should be "rows of the image" Page 479, line -3: "once to the rows of Figure 5.16b" should be "once to the columns of Figure 5.16b" Page 488, line -3: "was" should be "way". Page 490, line +15: "(239.5, 15.5, 192, 159, 127, 95, 63, 32)" should be "(239.5, 175.5, 111.0, 47.5, 15.5, 16.5, 16.0, 15.5)". Page 490, line -1: Matrix A_2 should be multiplied by the column (239.5, 175.5, 111.0, 47.5, 15.5, 16.5, 16.0, 15.5) Page 600, line+12: "the distortion values decrease" should be "the distortion values increase". Also the arrow in Figure 6.13 should be reversed (point toward the small values). Page 603, line+3: "In this type os search" should be "In this type of search". Page 763, Reference Manning (1998) has changed to: http://www.newmediarepublic.com/dvideo/compression/adv08.html Errors found by Michael W. Chou (UCLA) Page 121, Table 2.64: The renormalized values of C are wrong. The correct table is: Symbol C A Renor. A Renor. C ----------------------------------------------------------------- Initially 0 1 S1 (LPS) 0+1-0.1=0.9 0.1 0.8 0.9*2^3= 7.2 s2 (MPS) unchanged 7.2 0.8-0.1=0.7 1.4 7.2*2= 14.4 s3 (LPS) 14.4+1.4=0.1=15.7 0.1 0.8 15.7*2^3=125.6 s4 (MPS) unchanged 125.6 0.8-0.1=0.7 1.4 125.6*2=251.2 Table 2.64 Renormalization Added to Table 2.61. Errors found by Christopher M Brislawn While looking through the book, I noticed that you have added a section on the QM coder, which you state is used in several ISO standards. The situation is actually a bit more complicated than that: QM was the binary arithmetic coder in JPEG-1, but both JPEG-2000 and JBIG-2 use a different coder known as MQ (yes, how confusing). In fact, JPEG-2000 and JBIG-2 have coordinated their text so that they (supposedly; I haven't checked) use identical normative definitions of the MQ coder, to ensure that vendors won't have to build two inconsequentially inequivalent versions of it. You can get more details from the JPEG-JBIG website, http://www.jpeg.org/ ----------------------------------------------------------------- Error found by Khaled Asswad (Supelec, France) Page 302, line +15-16 "(the current rk in our example is 2)" should be "(the current rk in our example is 1)". ----------------------------------------------------------------- Error found by Jorgen Pramberg Page 43, the (only) display equation and the line following it should be: $$R=\left(-\sum_1^nP\log_2P\right)-\left(-\sum_1^nP_i\log_2P_i\right) =\log_2n+\sum_1^nP_i\log_2P_i.$$ The test for fully compressed data (no redundancy) is thus $\log_2n+\sum_1^nP_i\log_2P_i=0$. ----------------------------------------------------------------- Error found by Dharmendra Lingaiah Page 669, line -1: "80 ms" should be "8 ms". ----------------------------------------------------------------- Error found by Jason Corless Page 553, line -6: $|c_{i,j}|less than T$ should be $|c_{i,j}|\leq T$ Page 556, Figure 5.63. The table entry "HL1 4 Z 0" is missing. It should follow entry "HL1 3 Z 0". ------ Error found by Richard Pace Page 14, Table 1.1: The Braille code of the letter "H" should be x. xx .. ------ Errors found by Peter Dillinger Page 475, lines -9, -10: "the (high resolution) detail -1" should be "the (high resolution) detail 1". Page 475, line -9: "this can be expressed as (5,3)A_2=(4,-1)" should be "this can be expressed (up to a factor of \sqrt{2}) as (5,3)A_2=(4,1)" Page 499, line -9: "Notice how the output of the highpass filter $H_0$ of each stage" should be "Notice how the output of the lowpass filter $H_0$ of each stage". ------ Error found by M.S.Sriganesh and Bhupesh Jingar Page 58, line 3: $i-1$ should be $i$. Also in the table following, change $N-2$ to $N-1$ and $N-1$ to $N$. ------ Errors found by Cosmin Truta Page 512, figure 5.36 "Byelkin" should be "Beylkin". Page 53, section 2.4: The discussion of the Golomb code is incomplete and has a few errors. See revised text here. Page 550, figure 5.60. The box under 18 and 3 should be 2(1,1) 2(1,1) 1(1,1) -2(1,1) ------ Errors found by Alfred Fuchs Page 636 paragraph -3 has been rewritten to describe the Nyquist rate in terms of signal bandwidth rather than the highest frequency. It should read as follows: The solution to the sampling problem is to sample sound at a little over the Nyquist rate (page~466), which is twice the bandwidth of the sound (the bandwidth is the difference between its maximum and minimum frequencies). Thus, if a sound contains frequencies between 500~Hz and 2~kHz (the bandwidth of the human voice), it should be sampled at a little more than 3~kHz. (Previous editions of this book, as well as many other sources, erroneously state that the Nyquist rate is twice the maximum frequency. The author is indebted to Alfred Fuchs for clearing up this point. Another detail that's often ignored in the literature is that signal reconstruction is perfect only if something called the Nyquist-Shannon sampling theorem [Wikipedia~03] is used.) Such a sampling rate guarantees true reproduction of the sound. This is illustrated in Figure~7.3c, which shows 10 equally spaced samples taken over four periods. Notice that the samples do not have to be taken from the maxima or minima of the wave; they can come from any point. ------ Error found by Liyang HU Page 775, line 3: "A bi-level image, for example, consists of two bitplanes." Should be "A bi-level image, for example, consists of one bitplane." ------ "The road to wisdom? Well, its plain and simple to express; Its err, and err, and err again, but less, and less, and less." -- Piet Hein, Grooks.