##
**Optimization for computer vision. An introduction to core concepts and methods.**
*(English)*
Zbl 1280.90001

Advances in Computer Vision and Pattern Recognition. London: Springer (ISBN 978-1-4471-5282-8/hbk; 978-1-4471-5283-5/ebook). xi, 257 p. (2013).

The goal of this book is to provide a practical guide to optimization algorithms that are commonly used in computer vision applications. The book is quite different from standard textbooks on convex [S. Boyd and L. Vandenberghe, Convex optimization. Cambridge: Cambridge University Press (2004; Zbl 1058.90049)] and non-convex [J. Nocedal and S. J. Wright, Numerical optimization. 2nd ed. New York, NY: Springer (2006; Zbl 1104.65059)] optimization in terms of the selection of algorithms it discusses as well as in the way these algorithms are presented. The book focuses on optimization problems that commonly occur in computer vision, such as non-linear least squares problems, correspondence problems, deconvolution problems, and (discrete) energy minimization problems. The algorithms covered in the book are explained intuitively and are accompanied by pseudo-code descriptions and by example applications in computer vision. Mathematical details that are not of interest to the practitioner are generally omitted. This makes the book very well suited for computer-vision practitioners who are interested in identifying the most appropriate algorithms for their optimization problems and in implementing these algorithms.

The algorithms covered by the book include algorithms for first-order continuous optimization (stochastic gradient descent, conjugate gradients), non-linear least squares (Gauss-Newton, Levenberg-Marquardt), correspondence problems (ICP, RANSAC), and discrete energy minimization (graph cuts, dynamic programming). The examples of computer-vision models/problems in which these optimization methods are used are relevant models such as snakes, stereo matching, pictorial-structures models, deconvolution, and weakly supervised image segmentation.

Since many computer-vision textbooks and papers generally do not cover these optimization algorithms in much detail (they certainly do not commonly present pseudo-code, but instead refer to the optimization literature), this book is a nice reference that can be used in combination with the most computer-vision textbooks and papers.

The algorithms covered by the book include algorithms for first-order continuous optimization (stochastic gradient descent, conjugate gradients), non-linear least squares (Gauss-Newton, Levenberg-Marquardt), correspondence problems (ICP, RANSAC), and discrete energy minimization (graph cuts, dynamic programming). The examples of computer-vision models/problems in which these optimization methods are used are relevant models such as snakes, stereo matching, pictorial-structures models, deconvolution, and weakly supervised image segmentation.

Since many computer-vision textbooks and papers generally do not cover these optimization algorithms in much detail (they certainly do not commonly present pseudo-code, but instead refer to the optimization literature), this book is a nice reference that can be used in combination with the most computer-vision textbooks and papers.

Reviewer: Laurens van der Maaten (Delft)