A facial expression recognition algorithm with supervised orthogonal locality preserving projection

According to the characteristics of facial expression, this paper presents a facial expression recognition algorithm with supervised orthogonal locality preserving projection (SOLPP) based on combination of Gabor and local binary pattern (LBP) features. Because of the deficiency of traditional Gabor feature extraction method, an innovative feature extraction method based on Gabor local statistic information is proposed. Each Gabor wavelet representation of an image is divided into some sub-blocks. Then the mean value and standard deviation in each sub-block are calculated, and the statistics of all Gabor wavelet representations are connected as feature vector. Taking into account the effectiveness of LBP to extract local expression texture, we combine local statistic features of Gabor wavelets with LBP textural features as composite facial expression features. After utilizing SOLPP to reduce the feature dimension of composite features, the facial expression image is classified by nearest neighbor method. Experimental results on JAFFE database, CED-WYU (1.0) database and TFEID database indicate the proposed method has higher recognition rate compared with other methods.