Abstract: In order to improve the face recognition rate under the fewer training samples,this paper proposes an improved face recognition algorithm.Based on Gabor wavelet can characterize local texture feature of face image,this algorithm describes image by combining amplitude and phase information. extracting independent components through ICA method,using nearest neighbor classifier to classify the features and conduct test on the ORL face database.Experimental results show that this algorithm has a high recognition rate,Especially in the number of training samples of less cases,recognition rate is above 90%.