Medical Based On Improved LBP Operator And Sparse Expression On Facial Expression Recognition

As is known to all, facial expression involves expression of emotion and inner feelings.It is widely used in various fields including pattern recognition, machine vision, and intelligent control, etc. Therefore, the research of efficient facial expression recognition algorithm has great significance. The primary research of this thesis is to discuss the process of image feature extraction and image classification in facial expression recognition. The main contributions of this work are detailed as follows:(1) The related work of traditional facial expression recognition is firstly summarized,we then discuss the identification process of facial expression images in detail. Meanwhile,the thesis introduces the basic theory of compressed sensing, sparse representation, sensing matrix and the reconstruction optimization algorithm of sparse signal. In particular, the reconstruction optimization algorithm such as greedy pursuit algorithm and convex optimization algorithm are emphatically analyzed, respectively.(2) Due to the fact that the eigenvalue calculation of traditional LBP ignores the center pixel point of image, the operator of LBP based on partitioning strategy is proposed in this thesis. By this means, we calculate the characteristic value of the center pixel point and assign the highest weight. Meanwhile, the feature extraction based on the operator of C-LBP considers the influences of both partitioning strategy and feature dimensionalities. The proposed C-LBP method combined with GPRS classification obtains the 78.43% average recognition rate when the parameters of number of training sample and block strategy are respectively assigned to be 280 and 5*3. Note that the PCA dimensionality reduction method is used before the feature extraction and classification.(3) Owing to the sparse coefficients measured by the traditional OMP algorithm are relatively small or even to be negative, which leads to the misclassification. To solve this issue, an improved OMP algorithm is proposed in this thesis, which directly reduces the negative value of the sparse coefficients. Experimental results conducted on the JAFFE and the CK face databases demonstrate the effectiveness of the proposed method. Particularly, in the CK database, we can obtain the 80.30 average recognition rate using the improved OMP algorithm combined with PCA and the proposed C-LBP operator.