Automatic human face detection and recognition under non-uniform illumination

Abstract

A system for automatic human face detection and recognition is presented. The procedure consists of five steps: (1) the Haar wavelet transform, (2) facial edge detection, (3) symmetry axis detection, (4) face detection and (5) face recognition. Step 1 decomposes an input image, reducing image redundancy. Step 2 excludes non-facial areas using edge information, whereas Step 3 narrows down face areas further using gradient orientation. Step 4 restricts face-like areas by template matching. Finally, Step 5 determines the best face location in the face-like areas and identifies the face based on principal component analysis (PCA). The system shows a remarkably robust performance under non-uniform lighting conditions.