A novel 4-D perceptual quantization modeling for H.264 bit-rate control

Bit-rate control plays a major role in video coding and multimedia streaming. A well-designed bit-rate control mechanism can achieve fine visual qualities and avoid network congestion over a time-varying channel. This paper proposes an H.264 bit-rate control using a 4-D perceptual quantization modeling (PQrc), including two major encoding modules: the perceptual frame-level bit-allocation using a 1-D temporal pattern and the macroblock-level quantizer decision using a 3-D rate pattern. The temporal pattern is used to predict frame complexity and determine proper budget bits further. The rate pattern is depicted as a bit-complexity-quantization (B.C.Q.) model, in which a tangent slope of a B.C.Q. curve is a piece of unique information to find a proper quantizer. For newly generated video clips, the B.C.Q. model is updated continuously using a weighted least-square estimation. In comparison with the latest H.264 JM10.2, our experiment results show that the proposed PQrc can: 1) keep stable buffer fullness and 2) improve the SNR quality and control accuracy effectively.