Bayesian approach to image interpretation (Q2743237)

This book is an interesting contribution to Computer Vision, a science whose purpose is developing the theoretical and algorithmical basis through which computer-based automatic information extraction and analysis is feasible. More ambitiously, it consists of providing vision capability of the human eye to the computer. The main subject is image interpretation, that can be understood as the process of giving meanin to a 2D image by identifying and labelling significant objects or segments in the image. NEWLINENEWLINENEWLINEIn Chapter I, Overview, a brief and informative survey on the state-of-the-art is made, and a comprehensive bibliography is provided. Chapters II and III, Background and MRF Framework for Image Interpretation, give, in a precise and rigorous manner, the main theoretical concepts used in image interpretation: MRF, Hammersley-Clifford theorem, maximum entropy, a posteriori distribution and the basic ideas of simulated annealing. Looking at signals and analysing them at various scales of resolutions, is a promising new idea that is accomplished by the human vision system automatically. In Chapter IV, Bayesian Net Approach to Interpretation, is shown how MRF models allow the construction of a relatively simple, singly connected, two-layered Bayesian network for image interpretation. The definition and main properties of Bayesian networks are given, being the section on Bayesian networks for Gibbsian image interpretation of particular interest. Chapter V, Joint Segmentation and Image Interpretation, is, to my knowledge, where there are more contributions. A new method for image interpretation is proposed and detailed defined. An algorithm for its implementation is provided, and results are discussed on synthetic and real images. NEWLINENEWLINENEWLINEThe book ends with several appendices that provide a self-contained treatment of the subject. This book is a very interesting reading for those working on image interpretation, without the need of specialized knowledge on stochastic processes or statistics.