Abstract: A novel method for face recognition based on region is proposed to solve the problem of illumination variable. First the image is divided into four regions and the grayscale changes in the regional context is regarded as a measure of illumination affect the face display. The impact regions executes frequency normalization. Features is extracted by regional units and region recognition by support vector machine, then final result is achieved by recognition result fusion. The experiments show that the method can get good face recognition rate under illumination changes.

Key words: face recognition, variable illumination, region division, support vector machine, illumination normalization