Feature Fusion of Wavelet and LBP-GD for Facial Expression Recognition

Aiming at the problem that local binary pattern (LBP) cannot describe the change of texture direction, an LBP-gradient direction (LBP-GD) operator which combines gradient direction with LBP is proposed. LBP-GD operator not only keeps the advantages of LBP, but also describes texture direction in detail. Due to the difference of information contained in facial expression organs, an irregular dividing method is designed. The image is divided into 9 non-overlapping sub-blocks with different weighted values, then LBP-GD feature of each sub-block is extracted. Finally, the LBP-GD feature is weightily fused with low-frequency components obtained by lifting wavelet (LW) transform, and the K-nearest neighbor classifier is applied for expression classification. The effectiveness of this approach has been demonstrated on the JAFFE and Cohn-Kanade facial expression databases. Experimental results show that the proposed method achieves better performance than LW and LBP-GD feature alone.