A robust elastic and partial matching metric for face recognition

摘要: We present a robust elastic and partial matching metric for face recognition. To handle challenges such as pose, facial expression occlusion, we enable both by computing part based representation. In which N local image descriptors are extracted from densely sampled overlapping patches. then define distance where each descriptor in one is matched against its spatial neighborhood the other minimal recorded. For implicit matching, list of all distances sorted ascending order at αN-th position picked up final distance. The parameter 0 ≤ α 1 controls how much changes, or pixel degradations would allow. optimal values this new extensively studied identified with real-life photo collections. also reveal that filtering simple difference Gaussian brings significant robustness to lighting variations beats more utilized self-quotient image. Extensive evaluations on recognition benchmarks show our method leading competitive performance when compared state-of-the-art.