Exploiting parallelism in a vectorial fingerprint verification system

The BiOmega Automatic Fingerprint Verification System (AFVS) has been developed in the University of Puerto Rico. The initial implementation was done utilizing the Matlab development environment. The results in terms of accuracy, False Acceptance Rate (FAR) and False Rejection Rate (FRR), have been very positive. But the performance of the algorithms was too slow for a real-time application. After studying the inherent computing power necessary for image processing and analysis applications, it became clear that in the manipulation of images the same computations are always carried, but for different values of data. Because of this characteristic a symmetric multiprocessing computer was considered the ideal platform for achieving the desired performance for the BiOmega AFVS, 1.0 seconds. Two parallel architectures where used in this research effort, the commercially available Texas Instruments TMS320C80 and the SimPil array processor developed by the Georgia Institute of Technology. Both architectures offer multiple processing units in a single chip, and the advantage of local memory for each of the processing units. The focus of this research was to effectively translate the Matlab based BiOmega AFVS into the selected platforms to achieve the performance requirement of 1.0 seconds. (Abstract shortened by UMI.)...