Visual reconstruction for researchers

Computer vision poses a formidable information processing problem. Conceptually, visual information processing can be divided into three stages low, intermediate and high. The low and intermediate levels are critical in the visual processing hierarchy; it is here that information about the structure of images and surfaces is extracted and refined into a coherent representation which can drive subsequent highlevel object analysis tasks. Blake and Zisserman’s excellent monograph concerns itself with a class of inverse or reconstruction problem central to low and intermediate level visual processing. The authors’ theme is visual reconstruction which they characterize as reducing visual data to stable descriptions. Such descriptions must make explicit both the continuous regions and the disconti-nuity boundaries implicit in raw, noisy data. This amounts to a piecewise continuous reconstruction of the data. This book presents a detailed investigation into this problem using the powerful class of weak continuity constraints. The book is remarkably well written. Using readily accessible explanations and vivid examples, the authors present novel research including theoretical and empirical results gathered over the past five years. Formal analysis is postponed till well into the book and much of the supporting mathematics is relegated to the appendices. The result is a nicely balanced and selfcontained presentation which is at the same time concise and clear. The text begins by introducing the notion of piecewise continuous modelling and makes the important point that cooperative processing is the most attractive way to compute these models. Next, a selection of examples in image discontinuity detection, surface reconstruction and