The effect of low-bit rate JPEG2000 wavelet images on face authentication using advanced correlation filters

One approach in performing face recognition on mobile devices, due to its limited computational resources is to transmit the captured face images using the mobile phone connection and run the verification on a remote station. It may be necessary to transmit a compressed version of the image due to limitations in the communication bandwidth. This paper proposes the use of JPEG 2000 wavelet based image compression standard to compress the face images to very low bit-rates for transmission over the low-bandwidth communication channel. At the receiver end, the face images are reconstructed using a JPEG2000 decoder and fed into the verification algorithm. We explore the performance of advanced correlation filters like the minimum average correlation energy (MACE) filter and its variants using face images captured in different illumination conditions and encoded using different bit-rates under the JPEG2000 wavelet encoding standard. Evaluation of the performance of such filters is done using illumination variations from the CMU pose, illumination, and expression (PIE) face database. We demonstrate the tolerance of such filters to noisy versions of such images with variations in illumination condition.