Facial Expression Recognition Based on Curvelet Transform and Sparse Representation

Curvelet transform is a very effective multi-scale analysis tool with bandpass and directional, which has been proposed to optimize the limitations of the image edge feature extracted from wavelet transform. In the facial expression recognition based on Curvelet transform, low-frequency coefficients have the best performance on compressing and expressing the basic features of human face, detail layer coefficients mainly describe the local variation of facial expression, and high-frequency coefficients reflect the facial contour information. According to the facial feature region and the contribution of each Curvelet subband coefficients, we propose a facial expression recognition algorithm based on Curvelet transform and feature weighted fusion. The structural information representation of the image by Curvelet feature is enhanced. Meanwhile, it can also be used in combination with sparse representation based classification (SRC) which has powerful identification ability and robustness. Experiments on JAFFE database demonstrate that the algorithm we proposed can efficiently increase the capability of facial expression recognition and obtain high recognition accuracy.