Low-Complexity Analysis of Repetitive Regularities for Biometric Applications

Presented in this paper is a joint algorithm optimization and architecture design framework for analysis of repetitive regularities. Two closely coupled algorithm optimization techniques, referred to as the prime subspace periodicity transform (PSPT) and circular periodicity transform (CPT), are developed that significantly reduce computational complexity while enable the extraction of a wide spectrum of periodic features. The proposed PSPTCPT algorithms lead to a parallel and resource-sharing VLSI architecture. While most of the current systems rely on software solutions to performance feature extraction, the performance benefits rendered by the proposed framework show advantages in dealing with data-intensive computation for emerging applications in biometrics and bioinformatics. The explosive growth in database complexity combined with demand for fast analysis make the proposed framework a promising solution. Experimental results on an iris identification system demonstrate up to 99.2% accuracy and 24.6%− 41.9% reduction in computational complexity.