Real-time decision analysis - Algorithms, architectures and implementation

Abstract

Summary form only given, as follows. Decision analysis is the process of making an optimal decision (classification) based on extracted features as an input. Although an important task in supervised statistical pattern recognition, decision analysis is often the speed bottleneck of such a system. Most of the work on real-time pattern recognition was done in the area of feature extraction, and very little was done in the area of decision analysis. The design of a real-time decision analyzer that can operate at image sensor speeds is presented. The real-time performance is achieved by selecting a class of classifiers that is amenable to VLSI implementation and has considerable discriminatory power. A flexible system architecture for the decision analyzer is proposed. It can be tailored to particular user specifications and is based on a two-chip set as a building block. An application of the design to a low-level image segmentation system, called LISA, which is currently being built, is also reported.