Face tracking for H.264 encoded video sequences

In this paper we present a face detection and tracking-based scheme to enhance the perceptual quality of face regions in face-to-face teleconference applications using H.264. The tracking is realized using a particle filter. The weights of the particles are determined on the basis of intensity gradient near the edge of the ellipse, appearance difference and Bhattacharyya distance between weighted histograms. We assign smaller weights to particles which are farther away from the centers of detected faces. We employ a reliability factors of the utilized cues to control the noise variance as well as to supervise the number of particles. After classifying the macroblocks into face and non-face categories we perform a quantization. This requires some modifications in H.264 encoder but requires no modifications to the decoder.