Quantization scheme for high definition video coding based on node-cell pixel structure

An intra-only video coding algorithm based on a so called node-cell pixel coding structure has been proposed for high definition video by the authors. This novel coding structure divides pixels within one macro-block into node and cell pixels. The node pixels are first encoded using DCT, quantization, and entropy coding. The cell pixels are then interpolated using the reconstructed node pixels, and the corresponding residuals are encoded also using DCT, quantization, and entropy coding. Following this coding scheme, this paper further investigates the relationship of quantization parameters between node and cell pixels. To establish a model of this relationship, an extensive set of experiments were designed to perform both node and cell pixel encoding with different video sequences under different bit rates. The coding performance in terms of both bit-rate and distortion was measured for different combinations of the two quantization parameters. The optimal combinations, which bring the best coding performance, were identified and used to establish the mode. The experimental results show that the quantization parameter of cell pixels should be larger than the quantization parameter of node pixels to achieve a high coding performance. A new quantization scheme is designed based on this QP model and a gain up to 0.5 dB over the previous work is achieved.