Fractional low-order independent component analysis for face recognition robust to partial occlusion

As biometric vector may not follow the Gaussian distribution under complex light, pose and accessories, systems often yield unacceptable performance when subjected to impulsive, non-Gaussian noise. This paper adopts signal symmetric alpha stable distribution theory to construct fractional low-order independent component analysis algorithm (FLOD-ICA) and applied FLOD-ICA to solve the partial occlusion face recognition problem. Experiments verified that our proposed scheme is effective for face recognition under partial occlusion.