2D Object Detection and Recognition
Two important subproblems of computer vision are the detection and recognition of 2D objects in gray-level images. This book discusses the construction and training of models, computational approaches to efficient implementation, and parallel implementations in biologically plausible neural network architectures. The approach is based on statistical modeling and estimation, with an emphasis on simplicity, transparency, and computational efficiency.
The book describes a range of deformable template models, from coarse sparse models involving discrete, fast computations to more finely detailed models based on continuum formulations, involving intensive optimization. Each model is defined in terms of a subset of points on a reference grid (the template), a set of admissible instantiations of these points (deformations), and a statistical model for the data given a particular instantiation of the object present in the image. A recurring theme is a coarse to fine approach to the solution of vision problems. The book provides detailed descriptions of the algorithms used as well as the code, and the software and data sets are available on the Web.
About the Author
Yali Amit is Professor of Statistics and Computer Science at the University of Chicago.
"Modeling the human ability to identify objects in images has proved to be a significant challenge. While computer vision researchers have largely concentrated on the geometric aspects of the problem such as recognition under varying poses, researchers in statistics and machine learning typically have treated the problem as one of classifying feature vectors. In this important book, Yali Amit presents a novel synthesis of these strands of research. His approach to recognition based on learned configurations of sparse features provides a rare combination of efficient algorithms based on a solid statistical foundation. Amit's thorough and well-documented experimentation with examples ranging from medical images to handwritten digits should set a standard for the field. Highly recommended."--Jitendra Malik, Department of Computer Science, University of California at Berkeley
"The book develops a novel and elegant approach to the important problem of visual object recognition. The efficient and well motivated algorithms have fundamental theoretical as well as practical implications to the study of computer vision. The book will appeal to computer scientists as well as researchers modeling the functions of biological visual systems."--Shimon Ullman, The Weizmann Institute of Science, Israel