Facial Expression Recognition Based on Local
Vector Model, Yan Wang, Guoqing He · 2014
system. The LBP method is being researched and applied widely by researchers for its advantages such as strong performance of anti-jamming and convenient calculation. Compared with the AAM method and Gabor wavelet method, extract the texture features using LBP method does have obvious advantages. However, LBP approach ignored statistics for the change of direction, and cannot have a more detailed description for the strength and direction of the local texture changes, there are still some difficulties to extract the texture directly using the method to achieve a better classification. Thus, some researchers proposed the method of LDP [7], compared with LBP, the LDP method can extract more detailed texture feature to some extent, but the computing complexity is greatly higher than LBP. It is particularly necessary to find a method which can extract more detailed texture feature and less computing complexity for the establishment of the human-computer interaction system based on facial expression recognition. In this paper, we proposed a method of Local Vector Mode(LVM), in which we can get more detailed texture feature and lower computing complexity than LBP and LDP. The LVM is the vector of the local texture variation and have poor robustness for small deformation, so the Image Euclidean distance (IMED) [8] is introduced and embedded into the LVM method in order to extract more reasonable and effective expression features. Eventually, the method we proposed has the following advantages: Firstly, it can represents the strength and angles of local texture changes effectively. Secondly, it has a certain resistance noise ability and a stronger discernment. Thirdly, which has a better robustness for small deformation. Experiments on JAFFE facial expression databases with different resolution demonstrated that the method we proposed is better than the method of LBP and LDP.