Research On Perceptual Transform And Qualification Facing HD/UHD Video And Its Parallel Method

HEVC, the High Efficiency Video Coding standard, is the most recent joint video project of the ITU-T VCEG and ISO/DEC MPEG standardization organizations, working together in a partnership known as the Joint Collaborative Team on Video Coding (JCT-VC). HEVC has adopted many new technologies. However, HEVC still adopts the hybrid coding framework the same as H.264,in which intra-prediction and inter-prediction are used to eliminate the correlation of spatial redundancy and temporal redundancy, transform is used to eliminate the spatial redundancy and entropy coding is used to eliminate the statistical redundancy.As the final receiver of video, the human visual system (HVS) has perceptual redundancy because of its features. In order to quantify the perceptual redundancy, the researchers have done a lot of work. The Just-Noticeable-Distortion (JND) model is one of the widely accepted research. The JND model uses a threshold to qualify the visual perceptual redundancy. The change which is smaller than the threshold can not be noticed by human eyes. When the JND model is applied to video coding, because human eyes are not sensitive to the high frequency part which is below the threshold can not be encoded. Therefore, the rate can be reduced.In this paper, an improved Spatial-Temporal Just-Noticeable-Distortion (ST-JND) model is proposed. The ST-JND model includes spatial JND model and temporal JND model. The spatial JND model consists of the modeling of Space contrast sensitive function effect, the modeling of luminance masking effect and the modeling of texture masking effect. In this paper, according to the relationship between the background luminance brightness and the adaptive brightness threshold in DCT domain, we refine the lightness adaptive factor in the luminance masking effect to make the luminance masking model more accurate. At the same time, we apply to the Canny operator to edge detection and introduce the average edge weight coefficient to classify image to get more accurate texture masking effect model.We apply the improved ST-JND model to HEVC and use the structural similarity (SSIM) index instead of PSNR to evaluate the subjective video quality. The simulation results show that the improved ST-JND model can improve the coding efficiency largely, an average of about18.7%.In order to reduce the coding computation complexity of the DCT, quantization and entropy coding,an innovative algorithm is presented for detecting all-zero DCT coefficients blocks prior to DCT and quantization for all block-based video coding standards. This is A so-called early algorithm for detection for all-zero DCT coefficients blocks. A mathematic model for predetermination of all-zero DCT coefficients blocks is established based on the analysis of the features of the existing algorithms, Paseval energy conservation theorem and the characteristjcs of DCT formula and quantization. The algorithm is applied to HEVC video coder HM10.0, and experiment results show that the proposed algorithm can detect more all zero block and an average of about40%of the residual block could not be transformed and quanlified with an average loss of the image quality about0.0003. What’s more, this algorithm is not sensitive to the complexity of the video content. Besides, in order to further improve the transform speed, we apply the parallel algorithm of transform on the CUDA platform which is based on GPU and conduct a large number of experiments and analysis of the experimental results.