Combining Variation in the Bayesian Face Recognition

In the Bayesian face recognition algorithm, the similarity of two images is estimated on one intrapersonal variation subspace which is based on all variation training data. In fact, the intrapersonal variation is complicated and there are many factors which will lead to the variation. Different factor brings different influence to the face image. In the paper, we divide the intrapersonal variation subspace into four independent subspaces, and estimate the similarity of two images on each subspace, then combine the estimation results to give the final recognition results. Experiment results show that the algorithm is effective.