A low complexity and efficient face recognition approach in JPEG compressed domain using quantized coefficients

Computational and space complexities and storage space are amongst the most important issues in designing face recognition systems. A common method for storing images in face recognition systems is compressing images using JPEG standard. Usually, the compressed images are fully decompressed for recognition, so that the recognition process is done in decompressed domain. This procedure causes a high computational overhead. In this paper, we have studied the face recognition procedure using quantized coefficients in JPEG compressed domain for reducing the computational overhead caused by decompression process. In addition, in order to reduce the matching stage computational and space complexities, variance analysis and principle components analysis methods on quantized coefficients have been applied to reduce the dimension of images subspace. The experiments in this research have been done on four datasets of FERET database. Experimental results show that the proposed method outperforms existing methods in recognition rates, storage space, computational and space complexity aspects.