| With the rapid development of network multimedia technology,the application of highdefinition(HD)and ultra-high-definition(UHD)has been widely used,which generates a large amount of video data and brings great challenges to the storage and transmission of video.Based on this fact,the Joint Collaborative Team on Video Coding(JCT-VC)developed a new generation of video coding standard which is named High Efficiency Video Coding(HEVC)in 2013.Compared with the H.264/AVC,HEVC adopts a large number of advanced coding techniques and tools to double the compression efficiency of video coding.However,these techniques also result in intensive computational complexity,which seriously impedes the population and application of HEVC.Therefore,decreasing the coding complexity and optimizing the coding quality are two key issues in HEVC research.To meet the requirements of real-time UHD video coding,this paper focuses on the deep research of two key issues: the coding complexity and rate-distortion(RD)performance of HEVC,then we propose innovative algorithms to significantly decrease the coding complexity of HEVC and improve the perceptual rate-distortion performance.On this basis,the 4K UHD real-time video coding system based on HEVC is designed and implemented.The main contributions of this paper are summarized as follows:In order to reduce the complexity of HEVC intra coding,a fast intra mode decision algorithm based on logistic regression classification is proposed.In this paper,the partition of coding unit(CU)is formulated as a binary classification problem.By analyzing the characteristic of input data directly,a simple and efficient logistic regression classifier is used to early terminate CU splitting decision process so as to avoid traversal searches of the intra CU.In order to extract the features that are most relevant to the classification of the CU in the input data,the F-score evaluation method is adopted to select the features for different quantization parameters(QPs)and CU depth levels.The experimental results present that the algorithm reduces the coding complexity by 55.51% on average with 1.3% bitrate increase under the configuration of full I frame.To reduce the complexity of inter coding,a fast inter mode decision algorithm based on the Neyman-Pearson criterion is proposed,which consists of SKIP mode decision and fast CU size decision.Specifically,in this paper,the early mode decision is modeled as a binary classification problem,and the misclassification is divided into missed detection and incorrect detection.Then,the Neyman-Pearson decision criterion is adopted to minimize the missed detection rate under the premise of limiting the incorrect detection rate.A nonparametric density estimation scheme is developed to calculate the conditional probability distribution of the RD cost and other parameters,and the conditional probability distribution of the RD cost for different QPs and CU depth levels is periodically updated to improve the classification accuracy.Experimental results show that the algorithm can save 65% computational complexity on average with 1.29% bitrate increase.In addition,the algorithm also has the advantage of balancing RD performance and coding complexity by setting different values for the incorrect decision rate.Since the rate-distortion optimization(RDO)process used in HEVC neglects the perception characteristics of the human visual system,a perceptual RDO algorithm based on motion attention model and visual distortion sensitivity model is proposed in this paper.Different from general perceptual models,these two perceptual models make full use of motion vectors(MVs),transform coefficients,residuals and other information in the HEVC coding loop.These information not only reflects the motion characteristics of the object and the texture feature of the image,but also constrains the complexity increase of the model.In addition,as the MV is calculated at the minimum RD cost,sometimes it can not reflect the real MV.Therefore,an MV field refinement method based on maximum a posteriori estimation is proposed to improve the accuracy of this model.Then,according to the perceptual features extracted from these two perceptual models,the Lagrange multiplier and QP are adaptively adjusted to improve the perceptual quality of video coding.Aiming at the problem that panoramic video data is too large,a panoramic video coding method based on the gaze point is proposed.Due to the limitations of Head-Mounted Displays(HMD),panoramic video only displays the content of the viewport at any time,and human eyes usually focus on the area around the gaze point.Thus,in this paper,a gaze prediction model based on three-dimensional convolutional neural networks is proposed to predict the gaze point in panoramic video.The model considers both the video content related features and the history view path.Then,according to the predicted gaze point,an adaptive video coding method for panoramic video is proposed to improve the coding quality of the region of interest.The coding method adaptively adjusts the bit allocation and QP by combining the weighting factor of gaze point and the scaling factor from rectangular plane to spherical surface.Consequently,the bitrate of panoramic video coding is effectively reduced without degrading visual quality.We deeply investigate the encoding framework and computational complexity of x265,which is the widely used HEVC practical encoder.Then an efficient mode decision algorithm for x265 is proposed which includes decreasing the number of RDO times,early SKIP mode decision and fast intra mode decision etc.Meanwhile,due to the high computational complexity of the sample adaptive offset(SAO),the computational load between the predictive coding thread and the SAO processing thread is unbalanced,which leads to the low parallel efficiency of x265 encoding UHD video on multi-core processor(more than 12 cores).Therefore,this paper also proposes a fast SAO mode selection method to reduce the parallel waiting time,realizing realtime encoding of UHD video.In addition,based on the optimized encoder,a UHD real-time coding system is implemented,and an optimization strategy for system stability is proposed.The optimization strategy can adaptively adjust the coding speed and the bitrate according to the state of the input and output buffers. |
PDF Full Text Request