eBook Computer Vision download

Of the several computer vision textbooks that I haved owned and read, this book provides the best combination of introductory techniques with more advanced material in a very readable style.

The first two chapters are a very conversational overview of computer vision and image representation, but don't let this fool you. Starting in chapter three, the book becomes concise in presentation and in numerical examples. The authors starts out with the basics of binary image analysis which includes a very good discussion of image morphology. However, this is not an image processing book, so you should already be familiar with image processing on the same level as what is presented in Gonzales & Wood's "Digital Image Processing", which is my personal favorite among the various image processing texts. Next pattern recognition basics are discussed, including a section on neural networks that was clearer than anything I gleaned from Haykin's classic text on the subject. Next, the author moves into the realm of gray scale images by discussing the filtering and enhancing of images, which is similar to material in many image processing books. The basics of computer vision conclude with chapters on color, shading, and texture. Next, the book jumps into more advanced material that builds on the introductory material. For example, there are chapters on content-based image retrieval, a subject on which the author Linda Shapiro is conducting research at the University of Washington, and also on computing motion from 2D image sequences. Finally, the book tackles some 3D computer vision issues such as perceiving 3D from 2D images, object pose computation, and 3D models and matching using image "snakes". There are algorithms presented in pseudocode throughout this book, along with supporting mathematics, so the reader should have a good understanding of matrix algebra as well as calculus to really get the most from this book. The algorithms are concisely represented, and I had no trouble coding up a content-based image retrieval program based solely on the contents of this book. The pattern recognition chapter lacks a few details, and it might be helpful if the reader had a copy of Tom Mitchell's "Machine Learning", which parallels nicely with the pattern recognition chapter of Shapiro's book and is both complete and concise.

The book presents a nice complement to Image Processing, Analysis and Machine Vision (Image Processing, Analysis, and Machine Vision, 2nd ed., M. Sonka, V. Hlavac, and R. Boyle, 1998, IPAMV). As the difference in names implies, Computer Vision is not appropriate as an image processing textbook. It contains sufficient information on image processing to implement computer vision algorithms, but the focus of the book is on image analysis and high-level vision. The result is that the combination of IPAMV and Computer Vision cover the spectrum from intensive image processing and manipulation to high level analysis, object recognition and content-based image retrieval.
Computer Vision contains sixteen chapters that fall into roughly four categories: overview, 2-D CV topics, 3D CV topics, and special CV topics. Since it was written with the intent of reaching a broader audience than IPAMV, this book is appropriate as a primary text or reference for a wider variety of courses. For example, it would be appropriate for courses ranging from an introduction to imaging for non-scientists to a sophomore-junior elective to a first-year graduate seminar.
The overview chapters (chapters 1-4) include a summary of problems in CV, imaging and image representations, simple binary image analysis and a survey of pattern recognition concepts. The 2-D processing topics (chapters 3, 5-7, and 11) include thresholding and binary image analysis, filtering and enhancement, edge detection, Fourier Transforms, color, texture, segmentation, and 2-D matching and pose calculation. The 3-D computer vision topics (chapters 9-10, and 12-14) include motion detection and analysis, range image analysis, stereo, calibration, intrinsic image analysis and line labeling, shape from X, and camera models. The special topics (chapters 6-8, 15-16) include color and shading, texture, content-based retrieval, virtual reality, and a set of case studies of CV systems. Different combinations of these are appropriate for different types of courses.
In comparison with other texts, the coverage of color and shading in Computer Vision is the best available without consulting a color reference such as Fairchild's Color Appearance Models (described below). However, it still does not contain adequate coverage of physical models of reflection or color appearance. The texture chapter is comparable to Sonka et. al., and the CBIR and VR chapters are unique. It is these latter two areas that give Computer Vision a nice high-level flavor and provides a reference for these growing areas of CV.
Like IPAMV, Computer Vision contains a large number of example images, diagrams, and algorithms. The writing is clear and the mathematics--when it is necessary to present it--is complete and accessible. Since the book is designed with multiple audiences in mind, the heavy mathematical sections are flagged and the book can be used effectively with or without them.
Of particular interest to CV practitioners and students dealing with issues of calibration, chapter 13 contains a nice description of Roger Tsai's camera calibration algorithm, complete with an example. Note that Trucco and Verri (see below) also cover Tsai's calibration algorithm.
Overall, the choice between Computer Vision and IPAMV should be based on personal preference, the focus of your course, and the background of your students. IPAMV will be more accessible to engineers and contains more in-depth coverage of image processing techniques. Computer Vision is more accessible to computer scientists and covers a number of higher-level aspects of CV that are either not covered or briefly covered in IPAMV. In a number of areas--texture, stereo, motion, calibration, and segmentation--the two books are quite similar and the differences are mainly in style and emphasis.

Pro:
This book is really good. In simple, the book is written in English. It seems to be aimed at an entry-level CV student. Having some prior Image Processing or Computer Vision background will help you run through the book faster, although it doesn't seem to be required.

Shapiro makes sure you understand the concept behind the algorithm and then provides you the pseudo code rather than typing up some complicated C/C++ code.

Con:
There are alot of exercises in the chapter and they really help in testing your understanding. I only wish the author provided solutions to the exercises.

Shapiro's "Computer Vision" is an excellent book for someone looking for an introductory text in the field. The book is well structured and introduces fundamental concepts first, then uses these concepts to build on advanced approaches. The book assumes some knowledge of mathematics in linear algebra, calculus, and set theory, but does a good job of introducing the concepts before jumping into the math. Compared to other vision books this book is less focused on the math and more focused on conveying the concepts. Math intensive sections are noted up front in the table of contents.

The reason I rate this book 3 out of 5 stars is that the paperback is entirely in black and white. The original hardcover contained both color images and color plates. When these images are converted to B&W it defeats the purpose of having them as the reader cannot distinguish some of the effects occurring in the image which Shapiro is discussing (especially in sections discussing color image processing!). It is very disappointing that the publisher opted to go this route. If possible I recommend obtaining a hardcover which I give 5 out of 5 stars.