Unified pattern recognition and its application to handwriting recognition

Abstract

The goal of this thesis is the design and the implementation of a visual pattern recognition system based on the analysis of fundamental principles of human perception. The system must simultaneously be psychologically well founded and computationally tractable. The benefit of this strategy is two-fold ; first, it can be used as a model of certain information processing mechanisms in human visual perception. Second, it can be applied to the creation of computer systems, which have performance comparable to that of humans in notoriously difficult domains, e.g. speech perception, unconstrained cursive handwriting recognition, real world scene analysis, etc.