Content adaptive quantization scheme based on visual perception for video coding

This paper proposes a new quantization scheme which fully exploits the human visual perception and the local eye sensitivity characteristic in two layers. This new quantization scheme classifies macro-blocks into different categories and classifies frequency coefficients in the same block into different frequency strips, then adjusts the quantization parameter according to different categories or strips to reserve more sensitive details of video. Experiments results show that the new quantization scheme can achieve better subjective video quality with lower bitrates.