Abstract:The method of multi-features combined with Local Gabor Rank Pattern (LGRP) to extract facial expressional features is proposed in order to overcome complex Gabor filter which directly extracts the facial expressional features with some redundant information and the extraction is relatively single.First,the Gabor multi-directional and multi-scale features are extracted in images,and the LGRP is further coded to enhance the robustness and differentiation ability.Then,to extract the multi-orientation information and reduce the dimension of the features,two fusion rules are proposed to fuse the original Gabor features of the same scale into a single feature.Third,to extract facial expression features respectively,the Haar wavelet and Otsu threshold segmentation methods and three different features are fused through cascades.It can represent effectively local features and global features.Finally,the facial expressions are identified using Support Vector Machine (SVM).The average recognition rate of 94.36% is achieved in CK+ emotion databases.Compared with other methods,this method has achieved a good recognition rate and robustness.