Facial expression recognition under random block occlusion based on maximum likelihood estimation sparse representation

Occlusion is a big challenge for facial expression recognition and previous efforts are largely limited to a few occlusion types without considering the random characteristic of occlusion. Since the original sparse coding model actually assumes that the coding residual follows the Gaussian distribution, which may not be accurate enough to describe the coding errors in practice, so we propose a new scheme by modeling the sparse coding as a sparsely constrained robust regression problem in this paper. Firstly, in order to reduce the influence of occlusion for facial expression, the test facial expression image will be assigned different weights in all pixels. Secondly, because the occluded pixels should have lower weight values, hence, we update the weight through constant iterative until the convergence is achieved. Finally, the final sparse representation of test image can be calculated using the optimal weight matrix. And the class of test image can be determined by the minimal residual which associated with each class of training samples to the test image. The proposed method achieves better performance in JAFFE database and Cohn-Kanade database and experimental results show that it is robust to facial expression recognition under random block occlusion.