An effective Gabor-feature selection method for face recognition

Gabor feature extracted by Gabor wavelet is widely used in face recognition and proved efficiency. But the dimension of the Gabor feature vector is so high that makes it very hardly to compute and store the feature directly in practice. Some methods are proposed to reduce the dimension of Gabor feature in literature. A new method to select the key points in face image is proposed in this paper, and only Gabor features of these points are used in recognition. Except using the simple criterion in key points selecting, two algorithms in refining the selected key points are proposed also. Experimental results show that the proposed method is simply and effective not only for dimension reduction, but also obtains the better recognition performance.