Sparse Facial Expression Recognition Algorithm Based on Integrated Gabor Feature

Abstract

Facial expression recognition has been widely applied in the field of medicine and civil, and is a very active research area. Gabor wavelet transform is a classical and effective method of expression feature extraction, but the original facial expression images after the Gabor transform generate high dimension feature, which must be processed through effective feature fusion and selection, otherwise it will cause data redundancy. In this paper, in order to overcome the shortcoming of data redundancy of the traditional Gabor feature, sparse facial expression recognition algorithm based on integrated Gabor feature is proposed. Firstly, by means of two integration methods, mean fusion and differential binary encoding, the original Gabor feature images are integrated in a multi-scale and multi-angle way and 26 integrated Gabor feature images are obtained; then use feature selection method based on the facial expression recognition contribution coefficient, selecting 4 images from 26 integrated Gabor feature images as the final feature vector. Finally, the feature vector is fed to sparse representation classifier for facial expression recognition. Experimental results indicate that sparse facial expression recognition algorithm based on integrated Gabor feature can separate and express the facial expression features facial features effectively, and reduce dimension and present expression data compactly, meanwhile the expressions are classified correctly.