Expression recognition based on Scatter-Difference Matrix and Independent Component Analysis

Independent component analysis (ICA) is a basic method widely used in expression feature extraction and recognition. In this paper, combined with the characteristic of ICA, a novel method based on Scatter-Difference Matrix and Independent Component Analysis is presented. With the help of Scatter-Difference matrix, expression feature can be identified and classified effectively by ICA.Firstly, the difference between expression face matrix and neutral face matrix is computed to scatter-difference matrix. Then the whiten matrix can be gained. Finally, training and testing samples are projected into the independent space to get their features respectively and nearest neighbor distance (NND) rule is utilized in classification. Experimental were done on CED-WYU(1.0) and Japanese ART female JAFFE databases. Results show that correct recognition rate by the method is higher than that by 2DPCA, PCA- ICA and 2DPCA-ICA. Therefore, the method presented by this paper is valid in expression feature extraction and recognition.