Facial Expression Recognition Arithmetic Research

Facial expressions recognition is one of the most important subjects in the fields of Human-Computer Interaction and Information Processing, which has certain academic theory value and potential applications. It has great significance, for its breakthrough will radically change the relationship between people and the computer. At the same time, because of the complexity and specificity of the human emotion and the facial expression, it becomes a challenging subject.Learned many interrelated literatures and research papers concerning facial expression recognition of the domestic and international in recent years, some problems about feature extraction in facial expression recognition are analyzed. We do some researches deeply. The experimental results showed that the methods of facial expression recognition in this paper are reasonable and have both certain theory value and practical value. The research work in this paper mainly includes the following several respects:Firstly improved the weighted principal component analysis(WPCA )arithmetic. In this paper, the weighting principal component analysis (WPCA) arithmetic was improved and applied to the facial expression recognition system. According to the important information distribution characters of the facial expression, a new Double-Center and Double-Orientation-Scale weighting function to weight each feature of facial expression image are proposed. We called this arithmetic improved weighted principal component analysis (IWPCA) for simple. The experience results on the JAFFE database demonstrated the availability of IWPCA arithmetic.Secondly applied the two-dimension principal component analysis (2DPCA ) arithmetic to the facial expression recognition system. We compared and analyzed two single-direction 2DPCA , two-direction 2DPCA (2D-2DPCA[58]) and the PCA arithmetic under theoretical analysis. The experimental results on two facial expression databases show that two single-direction 2DPCA and 2D-2DPCA are better than PCA under both the person-dependent and person-independent conditions.Thirdly proposed a new arithmetic for feature extraction, called the two-dimension locality preserving projection (2DLPP) arithmetic. The 2DLPP not only avoids the singular value problem but also inherits the advantage of locality preserving projection(LPP). To reduce the dimension further, we combined the 2DLPP with the...