Shift Invariance based Feature Extraction and Weighted BPSO based Feature Selection for Enhanced Face Recognition

Abstract

This paper proposes a novel Feature Extraction technique called Shift Invariance based Feature Extraction (SIFE) and a novel Feature Selection algorithm, namely, Weighted Binary Particle Swarm Optimizer (WBPSO) for enhancing the performance of a Face Recognition (FR) system. SIFE uses Stationary Wavelet Transform (SWT) for combating pose variance, and WBPSO is used to achieve a reduced feature subset. In the Pre-processing stage, Entropy-based cropping and YCbCr segmentation based scale normalization is utilized to eliminate background in facial images. A mirrored testing scheme is proposed to improve the recognition rate of pose-variant images. The aforementioned FR system was tested upon ORL, CMUPIE and FERET face databases and the experiments were found to yield promising results.