Facial Expression Recognition Based On LTP And Sparse Representation

As a vital part of the technology of computer science and human-computer interaction, Facial Expression Recognition(FER) has drawn much attention recently. Facial recognition technology is a cross-discipline involving physiology、psychology、machine vision and many other research areas with a wide range of application value.This thesis expounds some theories of the feature extraction 、 dimension reduction and recognition for facial expression recognition, analyzes and summarizes some advantages and disadvantages of several important algorithms, improves some algorithms for some key issues. The primary content of this thesis are listed as follows:1. A novel method by fusing LTP and DCT features is proposed for facial expression recognition in this research to solve the problem that Local Ternary Pattern(LTP) can’t describe more global information of a image including image deformation and contour feature and Discrete Cosine Transform(DCT) can’t describe more local texture feature information of a image. Firstly, the human face is transformed with a single two-dimensional discrete haar wavelet transform and low frequency coefficients are extracted. LTP features are gotten in each block of the low frequency coefficients. Then, the primary information of the face image is centralized in a small number of DCT coefficients, which are used as the global feature of the face expression. Then, effectively fusing LTP and DCT features based on Fast Principal Component Analysis(Fast PCA), which are used as the effective feature of the face expression. Finally, SVM is used to distinguish each test facial expression images. The results show that the method is better than the single global or local features in the aspect of rate of recognition.2. A novel classification based on fusion feature and sparse representation is proposed for solving the problem of low recognition rate using sparse representation for facial expression recognition. Fusing LTP and DCT features can describe the facial expression image more completely than other features, for which a new facial expression recognition method based on sparse representation is formed by combining fusion feature and Sparse Representation Classifier(SRC). The results show that the method is better than the traditional classification based on SRC.