Research On Facial Expression Recognition Based On Local Feature

The facial expression contains luxuriant information about human behavior.It is important to analyze people 's emotion by computer to realize human-computer intercation. The most important factor that affects the recognition effect is the features extraction. However, there are some problems in the traditional feature extraction methods. The main contents of this paper are as follows :(1) Facial expression classification, recognition framework, classification accuracy and the typical methods of facial expression recognition and classification are studied. This paper analyze the method of the local binary pattern, the the local gradient coding(LGC) algorithm and the algorithm using LGC based on horizontal and diagonal prior principle(LGC-HD). The study found that the method of LBP considered only the gray of the central pixel and the neighboring pixels, but did not consider the distribution of the texture of the facial muscles, wrinkles and other local deformation in the expression information, the vertical information is useless in LGC algorithm and that we need to divide the image into more blocks to get detailed information. The improved method of feature extraction is proposed.(2) A feature extraction method based on LGC in 5×5 neighborhood(LGC-FN) is proposed. LGC-FN combines the advantages of the traditional LBP method, LGC and LGC-HD method and remove redundant information in the vertical direction. On the basis of extracting the information of the direction of horizontal and diagona in 3×3 neighborhood, information in the 5×5 neighborhood is added. But the LGC-FN method only considers the threshold value of 0, when the local muscle deformation is severe, the method is not stable. Therefore, the LGC-FN is improved.(3) A method based on TLGC-FN is proposed. First, we verify the robustness of the LGC-FN on the performance of facial expression recognition when the muscle released severe deformation. Second, we analyze the effect of threshold and image block on the recognition rate of facial expressions by setting different thresholds and blocks. Third, we need to determine the optimal threshold and block scheme.Finally, the experiment uses support vector machine(SVM) to classify facial expressions. Experiments show that our method can effectively improve facial expression recognition rate and the recognition performance of the violent deformation of facial expression.