A new preselection method for face recognition in JPEG domain based on face segmentation

Face recognition in JPEG compressed domain has turned into one of the important standpoints for reducing the computational overhead of decompression process without degradation in recognition accuracy. This domain needs some efficient methods for preselecting compressed coefficients and performing recognition process, which leads to improving the recognition rates and decreasing the computational complexities of the feature extraction methods. In this paper, a novel preselection method is proposed for these goals. For the first time a more efficient decompression process is presented, by performing face recognition in zigzag scanned coefficients. In the proposed method, the area of the face is segmented in prominent and non-important regions, and subsequently, the DC and the different number of lower frequency AC coefficients of each block are preselected, regarding the regions used for face recognition process. Experimental results show that the proposed method outperforms existing methods in recognition rate, as well as in time and space complexity aspects.