Adaptive Patch Alignment Based Local Binary Patterns for face recognition

This paper introduces a novel face recognition method based on Adaptive Patch Alignment Based Local Binary Patterns (APALBP). LBP is one of the most effective features to face recognition. However, the effectiveness of this feature greatly relies on face alignment, i.e., since LBP is in fact an image feature rather than face feature, pose difference will directly influence the recognition performance. APALBP is much more robust than original LBP. The novelty of this paper comes from 1) enrolling an adaptive patch alignment method, so that LBP feature can be directly applied on unaligned images; 2) putting forward a new solution to small sample problems in face recognition; 3) introducing a novel feature extraction which could be extended to general recognition problems. We present improved recognition results to demonstrate the effectiveness of our approach.