| In multimedia business,video is one of the most commonly used media.However,the amount of uncompressed video data is too large to be transmitted in the network with limited bandwidth.As one of the key technologies in video applications,video coding is dedicated to removing the redundant components in video data.With the rise of multimedia services based on the mobile devices,such as the live telecast,it puts forward higher and higher requirements on the real-time video encoding,the compression and anti-error performance.As an important part of video coding,intra coding is only based on its own spatial correlation and regard as the reference for subsequent inter coding.It is one of the key technologies to avoid the error propagation and improve the transmission quality in the poor network environment.In the latest video coding standard H.265/HEVC,the intra-coding achieves a significant improvement in terms of coding performance by using a large number of new coding tools,but also introduce high computational complexity for the encoder,hence,it can not be applied to most multimedia services directly.In this thesis,the optimization for intra coding of H.265/HEVC is studied for saving coding complexity while maintaining the coding efficiency.In this thesis,we focus on reducing the coding time by eliminating the redundant intra prediction modes and partition candidates,based on which,a fast intra-coding system for H.265/HEVC is designed.Additionally,we consider the parallelism of the algorithm,the stability of the coding speed and the complexity of the algorithm are considered,synthetically.In order to skip the redundant intra prediction modes in H.265/HEVC,an image texture extraction method with low computational complexity and high degree of parallelism is designed firstly.Then a texture distribution based fast intra mode prediction algorithm is implemented by using the distribution of eigenvalues,image texture and the optimal prediction mode.Experimental results have demostrated that compared with the X265 version 1.7,the average PSNR of the high coding quality reduced by 0.47 d B as well as the average bitrate increased by 1.5%,and the average PSNR in the low coding quality is decreased by 0.03 d B as well as the average bitrate increased by 7%.The time of intra coding is reduced by 80% at the same time.For skipping the redundant H.265/HEVC partition candidates,we first extract the coding parameters(image texture,quantization parameter and prediction unit size)which influence the value of Rate rate distortion cost(Rd Cost),then proposes a fast Rd Cost forecasting method by analyzing the numerical relationship between the coding parameters and Rd Cost.In order to avoid the PU partition candidates falls into local optimum caused by the Rdcost minimization principle.We propose a classification confidence evaluation model to evaluate the possibility of making the optimal partition by using the fast Rd Cost of partition candidates.Finally,the redundant partitions candidates is removed according to the value of probability.Experimental results have demostrated that compared with the X265 version 1.7,the average PSNR of the high coding quality reduced by 0.64 d B as well as the average bitrate increased by 1.96%,and the average PSNR in the low coding quality is decreased by 0.17 d B as well as the average bitrate increased by 8.35%.The time of intra coding is reduced by 72% at the same time.Based on the research mentioned above,the overall coding overall coding performance of the fast intra coding system with the fast algorithm described above is also tested.A large number of experimental results show that the encoding time of the system is 90% lower than that of the original intra coding of X265 version 1.7,while the average PSNR is reduced by 0.203 d B and the average bitrate is increased by 6.1%. |
PDF Full Text Request