ERRATA

Sonka-Hlavac-Boyle

Image Processing, Analysis, and Machine Vision

ISBN 0-495-08252-X

 Chapter/Section/Page Existing text Resolution p. 39 256 such cubes 512 such cubes p. 41 Sensors based on photo-emission principles explore the photoelectric effect Sensors based on photo-emission principles exploit the photoelectric effect p. 41 This phenomenon is explored in several technological elements This phenomenon is exploited in several technological elements p. 43 luminophore is exponential luminophore is described by a power-law function p. 44 The term “high-pass filter” when discussing losses within cables suggests that low-frequency signals will be rejected. It might be better to describe this as an “equalization” process that accentuates higher frequency components. Eq. 3.5 h(t)d\tau h(\tau)d\tau p. 53 The complex Fourier transform of a sinusoid consists of two spikes, not one as implied by the text. p. 53 The condition that \int|f(t)|dt is finite does not require an exponential decay in f(t), and would be met by a power-law decay faster than 1/|t| p.56, Eq. 3.23 argtan arctan p.56, Eq. 3.23 |F|^2 = (Re F^2) + (Im F^2) |F|^2 = (Re F)^2 + (Im F)^2 p.65 The number of possible DCT definitions seems to be half as many as stated – excluding the odd-extensions at both ends reduces the 16 initial options down by a factor of ¼ (exluding odd-odd, odd-even, even-odd, but leaving even-even). p.73 unitary matrix unit matrix p.73 It is unclear whether a Jordan canonical form can be achieved by similarity transformation of a matrix unless the given matrix is Hermitian (or symmetric). p.74 In the singular-value decomposition of a matrix, the matrix U is generally unitary, implying both row & column orthogonality, whereas V is only column orthogonal in general. p.81 The speed of acoustic waves is generally proportional to the square-root of the elastic modulus, whereas the text implies a direct proportionality to “the elastic properties of the medium” p.86 The origin of chromatic aberration is more strongly influenced by the propagation through the material of the lens, rather than through air that surrounds it or the boundary between the two. p.93, Eq.3.90 The scalar products are lacking either “.”s or transposes, e.g. n.L or n^{T}L p.99 It would be helpful to use the example of satellite imaging of an airport to give examples of each of the four levels of image representation. p.107 the image size is 2^{L−1}. the image size is 2^L. p.109 The reduction factor, \lambda appears to be defined by only an upper bound – it is unclear whether |Mi+1|/|Mi| is intended to be equal to \lambda or whether \lambda is some form of best-fit to this ratio. p.111 Multispectral images are more likely to be represented as vectors of matrices (or third-rank tensors) than by “binary matrices” p. 124, l. 5+6 Output image pixel g(i,j) … f(i,j) in the input image Output image pixel f(i,j) … g(i,j) in the input image p.134 The paragraph breaks around “A digital image is discrete in nature and so equations (5.32)...” seem to be in the wrong place. ??? p.139 The expression of the Laplacian of a Gaussian is not correctly expressed within polar coordinates. The correct expression for the Laplacian operator in 2D polar coordinates is \frac{1}{r} \frac{\partial}{\partial r} r \frac{\partial}{\partial r} + \frac{1}{r^2} \frac{\partial^2}{\partial\theta^2} p.139-140 This implies that equation 5.52 should contain a factor of (r^2/\sigma^2-2), and not (r^2/\sigma^2-1) as stated in the text. This is perhaps more straightforwardly obtained by calculating the Laplacian in Cartesian coordinates. p. 151 The parameter D0 as used in the Gaussian filter coincides with the dispersion sigma, not sigma^2. p. 151 As the Butterworth filter is normally used with a (power) frequency response that involves even powers of the frequency, it might be more usual to replace n with 2n. p. 159 The diagonal elements of the Harris matrix (equation 5.76) should involve $$\frac{\partial f}{\partial x}$$^2 and $$\frac{\partial f}{\partial y}$$^2 rather than second derivatives, to be consistent with equation 5.75. The immediately preceeding text should also not refer to the “second derivative”. p. 207 The bold-face on “main idea” is somewhat peculiar. p. 215 The discussion of the Hough transform jumps rather abruptly from identifying parametric curves to the detection of symmetry axes. It would be worth a few comments about how the Hough transform can be formulated to detect, e.g. a mirror symmetry. p. 238 The definition of the matching functions should probably include a scaling parameter alpha, e.g. C_3(u,v) &=   \frac{1}{\displaystyle 1+\sum_{(i,j)\in V} \alpha^2 \big(f(i+u,j+v) - h(i,j)\big)^2} to emphasise that the fact that the matching function is sensitive to the choice of intensity scale in f and h, and the number of pixels used in the summation. It might also be appropriate to renumber the definitions C_n à C_(n−1) so that C_2 involves squares of the pixel-differences. p.263 In the closing sentence before algorithm 7.2 it might be worth adding “both” before “these pixels being represented in the joint spatial-image domain”. p. 267 The use of bold-faces in the partial derivatives \partial \psi / \partial \mathbf{n}_R  etc implies that these derivatives are vectors or tensors (this convention often being used as a shorthand for \nabla \psi  etc). p. 268 The Euler-Lagrange condition in equation 7.21 is incomplete when the functional being minimized includes second-derivatives. A more complete condition also includes term − \frac{\d^2}{\d s^2}E_{{\bf v}_s}  (see LaTeX version of errata) p. 270 The first term in equation 7.24 should have y^2_y replaced by u^2_y. It might also be worth expressing the first term using delta notations (see LaTeX version of errata) p. 275 The paragraph opening “Let a curve moving in time t” might better refer to a “closed curve”. p. 278 Equation 7.40 is missing a “.” in the dot-product between\nabla \psi and \partial X / \partial t p. 280 Equation 7.44 implies that c is a second-rank tensor. It should probably refer to the divergence of |r_|−1r_ rather than its gradient, to be consistent with the expanded form involving _xx etc. (see LaTeX version of errata) p.282 The earlier definition of the interior of the contour has _ < 0 rather than _ > 0. To be consistent, all terms in equation 7.47 probably needs to change sign. p. 282 The derivation of μr. (|r_|−1r_) in equation 7.47 from “μ(Length of _)” in equation 7.46 is far from trivial, and worthy of at least a brief explanation. I suspect that this term may actually be … (see LaTeX version of errata) p. 309 The use of bold-face x etc in the first introduction of the 3D matrix I(x, y, z) doesn”t seem consistent with immediately following use of x as a (unit) vector. p. 338 The definition of the continuous and discrete versions of h(_x,_y) are not consistent if the same b(x, y) is used. It is likely that the continous version will involve the Dirac _-function, i.e. b(x, y) being infinite rather than unity on the contour so that the integral in equation 8.2 is non-zero. p. 339 The superscript in equation 8.6 should probably be int (math-italic) rather than “int”. p. 340 The coefficients a_0 and b_0 are probably not translationally or rotationally invariant, unlike the other coefficients. p. 340 Equation 8.14 should perhaps involve −uk rather than +uk so that the mean value of a(lk) is zero. p. 345 In the sentence “First, they change their shape less then their control polygon”, “then” should be “than”. p. 357 The definition of the “scale-invariant” quantity \eta_pq in equation 8.45 seems to imply that the scale alpha  is known a priori. Given that the quantity μpq scales as _−(p+q+2) under x, y ! _x, _y, it would seem more likely that _pq should be defined as: _pq = μpq (μ00)(p+q)/2+1 i.e. using μ rather than μ0. (see LaTeX version of errata) p. 360 In algorithm 8.6, the definition of v might be more clearly expressed as Pq − Pk, rather than using an expression suggestive of the product of two vectors. p. 368 The regions being described in Figure 8.34 a&b are unclear. It would be helpful to shade those regions, or describe their shapes (e.g. “between the triangular boundaries in (a)”). p. 370 After the reference to Bookstein 1991, there is a double full-stop before the sentence starting “If such a landmark model”. p. 460 we seek d b we seek \delta b p. 462, algorithm 10.4, item 4 Ends with “and halt” Chapter 10, Fig. 10.26, p. 491 Alg. 10.9 referenced Alg. 10.10 shall be referenced Chapter 10, eq. 10.28, p. 494 extra right “)” Remove right “)” Chapter 15, p. 724, after eq. 15.9 \sigma_x and \sigma_y are defined as variances, not standard deviations redefine p. 771 … this other … this and other Index R-table missing Add R-table