Perception-based video coding with human faces detection and enhancement in H.264/AVC systems

A perception-based video coding algorithm for H.264/AVC systems is proposed in this paper. Obviously, the areas containing human faces and that with larger motion activities are or most concern by human visual perception system in which more bit resources should be allocated so that better visual quality can be exhibited. To do this, those macroblocks (MBs) containing human faces are determined by using an adaptive boosting (AdaBoost) classifier. To pick out macroblocks with larger motion activities, comparisons on the motion intensity of individual macroblocks with respect to the average motion intensity in previous video frame are performed. After that, a proposed dynamic quantization parameter (QP) adjustment strategy is applied based on a predefined coding priority so that limited resources can be concentrated to those regions of interest (ROI) under lower target bit-rate applications. Experimental results show that a very good subjective visual quality as well as objective PSNR can be obtained with the proposed approach when compared with the standard JM software and existing state-of-the-art algorithm on those regions of interest under the same target bit-rate constraint.