Face recognition base on low dimension Gabor feature using direct fractional-step LDA

A novel approach of Gabor feature based face recognition approach is presented in this paper. Gabor wavelet is a very powerful tool to analyze the texture of an image. Due to its characteristic of being insensitive to the varying illumination condition, it also has been widely used in face recognition task. However, the redundancy of Gabor feature limited its application. In this paper, a feature mask is applied for the sub-sampling of the original magnitude feature of Gabor transform. What´s more, a lately algorithm of discrimination, Direct Fractional-Step LDA, is adopted instead of traditional method of PCA+LDA. It preserved more important discriminate information during reducing dimensions. The experiment results indicate that the performance is superior to single DF-LDA and the method based on low-dimensional Gabor feature with PCA+LDA.