Matching structural and implementational models in the specification of image classifiers

摘要

While a direct relationship between problem specification/processing algorithm and implementational structure can often be found, a more effective match between high-level specification and the target hardware can often be obtained by considering an intermediate computational model which integrates the algorithmic structure and the implementational structure. This paper discusses the parallel implementation of pattern recognition algorithms applied to image analysis tasks. The investigation demonstrates how an intermediate level of modelling the classification procedure can lead to the specification of a novel structural reconfiguring of the basic algorithms which achieves improved performance both in terms of attainable error-rates and in terms of processing speeds.