Implementation of a training set parallel algorithm for an automated fingerprint image comparison system

Abstract

This paper addresses the problem of developing an efficient training set parallel algorithm (TSPA) for the training procedure of a neural network based fingerprint image comparison (FIC) system. The target architecture is assumed to be a coarse-grain distributed memory parallel architecture. Theoretical analysis and experimental results show that TSPA achieves almost linear speedup performance. This parallel algorithm is applicable to ANN training in general and is not dependent on the ANN architecture. However, TSP is amenable to a slow convergence rate. In order to reduce this effect, a modified TSPA using weighted contributions of synaptic connections is proposed. Experimental results show that this algorithm provides a fast convergence rate, while keeping the high speedup performance obtained. The above algorithms are implemented and tested on a 32-node CM-5.