Browse Books

Reviewer: David Marshall

Machine learning is a current topic of great interest to those working on pattern recognition and, more generally, to the computer science community. Over the past five to 10 years, machine learning techniques have matured and now permeate many research fields. Computer vision is one field where machine learning has substantially changed the practice in many subfields. One important example is face recognition, which has posed a fundamental challenge to the discipline for many years. Machine learning is now a cornerstone of contemporary computer vision. This book should, therefore, be of interest to anybody involved in computer vision or image and video analysis, as it presents many challenging scenarios to the machine learning community. Researchers in machine learning have for many years looked to such data to test and develop new algorithms. It is surprising that even though machine learning has been influential in computer vision for several years, there are relatively few books that directly address these topics together. A good text from 2005 is the most prominent [1], as it is the only real textbook on the subject. It provides a good theoretical background, but its example applications may be starting to seem dated. The book reviewed here is not really a textbook, however. It is a collection of papers covering some of the talks and tutorials from the last few editions of the annual International Computer Vision Summer School (ICVSS). According to the preface, courses at ICVSS "are delivered by world-renowned experts in the field, from both academia and industry, and cover both theoretical and practical aspects of real computer vision problems." As one would expect, the summer school is well known in the field and is probably the highest quality event of its type currently running. There is no doubt about the quality of the research presented in this book. It is a collection of current research snapshots by many of the leading computer vision scientists worldwide, and is best described as a survey of the current state of the art in computer science and machine learning. However, as is the case with such texts, it may become dated quite rapidly. The book is not without a few faults. Several chapters are a little too brief in technical detail, and in chapter 1, the print quality of some of the images and graphs is not as crisp as it should be. That said, there are a few excellent chapters in the book. Chapter 1, "Throwing Down the Visual Intelligence Gauntlet," written by leading researchers from MIT, sets the scene for developing modern intelligent machines. The authors describe a neuromorphic approach and discuss future research directions in the quest for visual intelligence. As a review paper, it is a little brief, but it supplies plenty of references for follow up. This chapter sets the context for the rest of the book quite well. The second chapter, "Actionable Information in Vision," by Stefano Soatto, addresses the challenge of extracting high-level visual information that can support vision-based decision and control tasks. This offers a contrasting approach to the more standard bottom-up information theoretic approaches to image analysis. It is a thoroughly interesting paper with good depth. Chapter 5, "Real-Time Human Pose Recognition in Parts from a Single Depth Image," was written mainly by researchers from Microsoft Research. This might well be the headline-grabbing paper in this book. It describes the operation of the famous Kinect Xbox console. The Kinect has revolutionized 3D interactive game playing. Enabling machine learning from real and synthetic 3D poses is the game changer in developing a commercially functional computer recognition system. Although a little brief, this chapter covers its main concepts well. The final chapter presents a case study in the daunting world of autonomous road vehicles (daunting, at least, for human drivers who mistrust current computer technology!). Researchers at the VISLAB in Italy designed a passenger car that recently logged more than 13,000 kilometers, driving from Parma, Italy to Shanghai, China, in three months. This paper describes the technical developments used to achieve this impressive feat. Other chapters deal with color-invariant object recognition, 3D object registration and recognition from point clouds, object tracking, and recognition, all important topics in computer vision. In summary, this book presents a snapshot of key research in the areas of computer vision and machine learning. On this level, the book succeeds, with many first-class papers. I recommend the book to practitioners in the field, as well as those pursuing PhD-level studies. It is not suitable for use as a textbook because it lacks a general introduction and basic theory is missing or only touched upon in some key chapters. The need for a basic graduate-level text on these topics remains. However, this book does go some of the way toward addressing the unusual lack of dedicated texts in these maturing and increasingly crosscutting areas. Online Computing Reviews Service