Ultra-Low Bit Rate Facial Coding Hybrid Model Based on Saliency Detection
Wei Zeng, Mingqiang Yang, Zhenxing Cui · Journal of Image and Graphics · 2015
Aiming at getting high quality image sequences in real-time video chat applications at very low bit rate, we propose an improved model-based video coding system in this paper. Instead of detecting human faces with conventional ways, we detect salient regions in the frames using Boolean Map based Saliency (BMS) method to locate the faces. After facial feature extraction by AAM, we transmit the DPCM of the eyes and mouth directly instead of time-consuming facial expression estimation. The new approach we proposed is more suitable for real time and the simulation below indicates the proposed algorithm is effective and feasible.