Towards Face Recognition in JPEG2000 Compressed Domain

The goal of this research is to examine the feasibility of implementing face recognition algorithms directly into JPEG2000 compressed domain. By exploring some standard face recognition algorithms' behavior when fed with standard JPEG2000 DWT coefficients (CDF 9/7 wavelet), we will prove that recognition can be done directly in JPEG2000 compressed domain, avoiding inverse discrete wavelet transform (IDWT). Such an approach would consequently enable the use of compressed images in recognition purposes, thus reducing both computational time and storage requirements. Our experimental results indicate that face recognition performance in JPEG2000 compressed domain is comparable, or even better in some cases, than face recognition performance in pixel domain.