It is rare to get a book that you don’t want to put down, but also one you can’t keep holding. At almost 900 pages, this is not a book that anyone can hold in hand and read for long; however, it so exhaustively covers this important topic of computer vision that readers will want to. With the resurgent interest in artificial intelligence (AI) and popularity of videos and cameras, computer vision has become a very popular topic. There are so many use cases for computer vision technologies, from image/video search to surveillance analytics. The book is indeed timely and relevant.

The opening chapter is an introduction to the rest of the book, pointing out the big picture. The remaining chapters are divided into five parts. Each chapter ends with an interesting bibliographical and historical section that takes readers through the major developments in the field and links to relevant literature. This is much better than the flat list of references that is usually found. Each chapter begins with an appetizer, giving readers a bird’s-eye view and highlighting the key things one can expect to encounter.

Part 1, aptly titled “Low-Level Vision,” looks at the basics of images and image processing. Image representation, filtering, morphological operations, thresholding, edge detection, and dealing with special features like corners, interest points, and texture analysis are covered in six chapters.

Part 2 is on “Intermediate-Level Vision,” as can be expected. Between the low-level details and the high-level picture, this layer provides the bridge. Chapters 8 to 12 cover binary shape analysis, boundary pattern analysis, the detection of line/circle/ellipses, the generalized Hough transform, and object segmentation and shape models.

Part 3 switches focus to machine learning and deep learning, which is a very critical ingredient of higher level vision processing today. Given our poor understanding of how human vision identifies patterns such as the human face, trees, and so on, we are very dependent on machine learning. The difficulty of identifying meaningful features drives one to explore deep learning, and this area has become one of the success stories. Machine learning is introduced gradually from basic classification concepts (chapter 13) to probabilistic methods (chapter 14) to deep learning (chapter 15). Major topics such as Bayesian networks, nearest neighbor, neural networks, and support vector machines are all covered briefly.

Part 4 starts to look at real vision issues, moving to 3D handling and motion. Chapter 16 introduces the 3D world, covering representational and processing challenges (for example, stereo vision, object recognition, and so on). The following four chapters are on the perspective n-point problem, invariants and perspective, image transformation and camera calibration, and motion. The treatments are relatively brief, with the entire section coming to less than 150 pages.

Part 5 is about putting computer vision to work and looks at specific application use cases. Face recognition, one of the most celebrated successes of deep learning, is the focus of chapter 21. Surveillance is covered in chapter 22 and in-vehicle vision systems in chapter 23--this is key to another hot topic: autonomous vehicles. Finally, chapter 24 is a collection of thoughts on field spanning representation, the role of deep learning, Moore’s law, and so on.

There are four appendices: “Robust Statistics” (an emerging field to make algorithms robust in the presence of outliers and noise), “The Sampling Theorem,” “The Representation of Color,” and “Sampling From Distributions.”

The fact that we are looking at the fifth edition of a book is itself a commentary on its quality and success. Well deserved, I must say, for its extensive coverage of important topics, from the fundamentals to the current state of the art. I like the organization of the book and the in-depth treatment of the current hot topic of deep learning. On the whole, if you are in any way interested in computer vision, you can’t live without this one. Topics such as this do require some mathematics, so the scattering of scary formulae here and there should come as no surprise. One improvement I would like to see is a larger collection of end-of-chapter problems for reflection, understanding, and pushing ahead; in this edition, typically only two to four problems are included in each chapter.