Audio-visual recognition system insusceptible to illumination variation over internet protocol

In this paper, we present an audio-visual recognition system which is insusceptible to illumination variation over internet protocol. First, the multiband feature fusion method is proposed for face recognition under varying illumination. The wavelet packet transform decomposes an image into various frequency subbands. We show how to select a set of subbands that are invariant to illumination variations by using statistical method and modified Euclidean based method. More specifically, there exist a set of wavelet subbands that carry facial features which provide an effective representation for face recognition under wide range of lighting conditions. Histogram equalization is then applied on these subbands to enhance the contrast of the features. The recognition performance of the proposed method is validated on some standard data sets and high recognition accuracy is achieved. Then the audio-visual recognition system over internet protocol is developed. The compression and packet loss effects of sending the audio and video data over internet protocol on recognition performance are investigated.