Expression recognition based on manifold learning under occlusion condition

An expression recognition method for occluded images is proposed in this paper.Firstly,tested images with occluded area are reconstructed using principal component analysis(PCA).Then,occluded area is detected using normal distribution theory.Tested images are embedded into manifold space based on partial similarity between images.Finally,support vector machine(SVM) is used to classify tested expression images.This method can eliminate the influence of occlusion detection error on expression recognition.It is robust to the expression recognition of occluded images.The method is testified by the experimental results on Cohn-Kanade and JAFFE face database,which has strong robustness,high recognition rate and efficiency.