Abstract

This paper presents a novel idea for implementing face recognition system in compressed domain. A major advantage of the proposed approach is the fact that face recognition systems can directly work with JPEG and JPEG2000 compressed images, i.e. it uses directly the entropy points provided by the compression standards as input without any necessity of completely decompressing the image before recognition. The Kd-tree technique is used in the proposed approach for the matching of the images. This algorithm shows improvement in reducing the computational time of the overall approach. This proposed method significantly improves the recognition rates while greatly reducing computational time and storage requirements.