Computer Vision: Algorithms and Applications (Texts in Computer Science)

A great introduction to Computer Vision, a nice review of the history of Computer Vision, and an enlightening survey of current and ongoing research. Richard Szeliski is a great teacher, at the top of his game, who gives motivation for the problems we may need to solve using Computer Vision. The algorithms are not provided as software code, but as descriptions with plenty of mathematical equations, references to papers, and copious diagrams and color photos. An enjoyable read, there is something for everyone interested in Computer Vision in this book. But although it is very broad, packing 700 textbook-sized pages with information, it does not always go very deep. And there is no source code. So you're on your own if you want to turn the discussed algorithms into working code. It's apparently intended to be used as a textbook, as there are questions at the end of each section. So reading each of the relevant papers and producing working software algorithms is left up to the reader. The example applications are motivating and there are a huge number of paper references (the footnotes section takes up 100 pages at the end of the book, just before the relatively-small 20-page index.)

For anyone looking for comprehensive coverage of all the fundamentals of computer vision, this is the book for you. The style of the book is that he gives you the general concept of a method and the required equations, and then provides you with the title of the paper it is sourced from (which are almost all available online as PDFs) in case you want more detail. Because of this approach, he has managed to fit into 800 pages what it could easily have taken 2000 to explain in fine detail. Don't take this as the book being too vague, because I have yet to need to refer to the papers to understand a concept (although it has been useful when I wanted more information into *how* the result was arrived at). It does help if you have a baseline understanding of how to do general operations on images (e.g. array manipulation in code) and whatnot, but a beginner could still use this book with a little more effort. There's no book that covers this breadth of information on computer vision, so I can't recommend it more highly.

I'm returning to CV after a long absence, and this book is perfect for me. I've done three deep topic dives so far, and the book's coverage of each (with additional pointers into the literature) is spot on. The author has generously provided a PDF version online (search for the book title). Still, the physical version of the book is really nice, with large format pages on good paper. Since the book has already paid for itself in time savings, I'm happy to have made the purchase.

The book acts as a good high level introduction to various significant sub-fields inside of computer vision. It is also one of the more up to date books (as of 2012) discussing more recent advances. However, because it is so high level and attempts to cover so much information, it is not a good book to try to learn from alone and provides no practical information on implementation details or problems. The best way to use the book, in my opinion, is to skim through it and learn the keywords to search for and use the references as a starting point. If you are a self learner like myself, one of the more frustrating problems is when you search for the wrong keywords and can't find the material. On the whole the writing quality is good in terms of clarity and insight. There are some spots that I felt that the book did little more than restate what was said by other authors of highly cited papers. While this is not always a bad thing, there were times that I disagreed with statements due to various practical considerations. Yes the original author was technically correct, but in the years since publication very few people actually do that since it's too computationally expensive or turned out to be less stable than advertised. The layout and organization of the book is well done and contains many full color pictures. For a new text book it is also very reasonably priced. I suspect that it would be more expensive to print the book's PDF out in color rather than buying it! I bought the book instead of just skimming through the older drafts (available online from the author's website) primarily because I prefer printed books and to support the authors publication approach. My rating is 4 stars based upon it being a high level introduction. As mentioned before, if you want a practical book that goes into how to implement all the techniques it discusses and issues that will arise, look elsewhere.

It's impossible to cover the whole topic in one book. This one has the main ideas, some explanations and examples, and a lot of external references. It's not an easy book (as I was hoping;)). You will need other sources to understand where the formulas came from, and what do they actually mean. As a result you will have a good understanding. It's better to start with this rather than with product oriented books (like OpenCV). These should go after, IMHO. PS: Author provides a free beta PDF (very smart idea), makes sense to check it first. I realized that was what I was looking for, and bot the book.

I was introduced to this book in my undergrad course on Computer Vision. As it was only a semester long course was not able to thoroughly dive into and understand all the topics as much as I would like to. As I am currently getting ready to pursue a graduate degree focusing on machine learning and computer vision I feel this book is a great help towards having a deeper understanding of these topics.

to really understand vision you need to understand projective geometry and how to project a scene from 3-d to a 2-d screen. This book covers various pre-deep learning techniques. I especially loved the sections on photogrammetry which is about going from a 2-d image to a 3-d scene which lets you do stuff like import your furniture into VR so that you don’t bump into stuff.

Disclaimer: I'm just a senior engineering undergrad who has had some relevant experiences read up all over the web and recently started taking a CV course I just started to do some sensor fusion and calibration stuffs and have been scratching my head trying to look up references on technical details all over the web, but unfortunately it's been more difficult than I thought. Most online stuffs aren't very organized or designed for somewhat beginner, barely touches any technical details required to understand inner workings of sensors and maybe do implementation from scratch. As soon as I got this book, I flipped to a random page, and it's "damn I've been looking up on this all over the web and couldn't understand anything beyond the surface of what seems easy but doesn't provide enough detail to get me started on implementation or even just at least connect stuffs together" and this book is everything I wished for! And damn it's beautifully printed!

Can all cvpr authors answer any question from this book? Else they shuld be........

Covers all novel methods in computer vision, except deep learning which started after the book was published. I very much recommend to use the book and maybe additional papers if deep learning is of interest.

Excellent book as I expected.

I bought this book in order to have a more thorough understanding of the algorithms I was covering in my Computer Vision class. I was told this was 'the bible' in CV so was very excited to read it. I could not have been more disappointed. This book simply reviews all the techniques that exist in the field without going in any detail for most of them (from the couple of chapters I have read). It only covers very general aspects of the algorithms and summarises entire techniques in 1 or 2 paragraphs (eg. SIFT descriptor). Furthermore, it is impossible to read a paragraph without being constantly interrupted by references. I bought this book to learn the intuition and details of the techniques used in CV, not to get references to the authors that originally developed the algorithms. If I want to have a look at the original papers or the work of their authors, I can simply google it. Although I have been severely criticising this book in the last two paragraphs, I want to remind the reader that I have only read 2 to 3 chapters and that it may not be representative of the entire book.