Face recognition based on regression analysis using frequency features

Linear regression analysis is an important method in face recognition. It establishes a new framework for feature extraction and small sample size (SSS) problem. However, in real world, the accuracy is susceptible to occlusion, illumination and varying pose. In this paper, we present a novel method using frequency features of image to recognize different individual face. According to robustness of two dimensional discrete cosine transform (2DDCT), it is used to transform the images signals form spatial domain into frequency domain aiming to reduce the noise effects of original images. The coefficients, which are maintained by threshold, will be considered as the features. In addition, We make a decision to classify the face image. Based on the properties of 2DDCT, the major features of each image are concentrated in upper left corner and this region is treated as a module. So, the fusion information combines major features and module. Experiments on two face databases such as OLR, YALE show the promising performance of the proposed technique.