Facial Expression Recognition Based On Adaptive Weighted LBP And Collaborative Representation

Facial expression is an un-natural language in face-to-face communication, the information represented in the expression have played a vital role in the topic of human-computer interaction and emotional research, and researching the topic have also played a important role in artificial intelligence field. This paper mainly adopted the adaptive weighted local binary patterns and collaborative representation based classification algorithms to research the face expression recognition. Feature extraction, feature dimension reduction and facial expression classification are three main aspects of facial expression recognition. This paper mainly studied the above three aspects in expression recognition field, the research work of this paper is as follows:(1) As to this problem that the original LBP operator can’t express the mutual collocation between local features and can’t distinguish the texture information of different regions, we proposed an adaptive weighted LBP method to extract the feature of facial expression images in this paper. The main steps are as follows: firstly, normalized the expression images’ size; secondly,divided the image into uniform blocks considering the size of the image and image block scheme;then extracted the blocks feature using the uniform patterns LBP and calculated each sub block histogram entropy; finally, weighted the expression image feature histogram using each sub block histogram entropy value.(2) In order to meet the classification condition in the following classification and study the relationship between the feature dimension and the recognition rate, it is necessary to reduce the feature dimension. In this paper, the principal component analysis is used to reduce the feature dimension of the weighted image. Because sparse representation classifier and collaboration representation based classifier needed to meet the condition that training sample dictionary’s dimension is less than the number of training samples. At the same time, in the final experiment, the relationship between feature dimension and recognition rate was also studied.(3) As to the problem that the classified time is too long, this paper proposed a classification method based on Collaborative representations to accomplish the classification of facial expression recognition. Because Collaborative representations classification method is mainly evolved from sparse representation methods, in this paper first elaborated sparse representation classification principle, then compared two types of algorithms from the two aspects of theory and experiment and confirmed the effectiveness of collaborative classification. In the experiments, the average recognition rate of facial expression classification is about 97%, and the average recognition time is about 0.1s.