Research On Rate Distortion Optimal Coding And Rate Control Method Based On Hybrid Distortion Measures

With the rapid development of large-capacity storage devices,high-speed communication networks and high-performance computing technologies,interactive visual media communications have become more and more popular,and people also have higher requirements for the quality of video experience.But the data volume of video is usually large and the growth rate is much faster than the expansion of network bandwidth.Therefore,the efficient encoding of video is essential,that is to provide users with higher quality visual experience under limited bandwidth conditions.Rate distortion optimization(RDO)is one of the main methods to improve coding performance in video coding.In H.265/HEVC,sum of squared error(SSE)is used as the distortion measure in the RDO process.The fidelity of data refers to the similarity of pixel values between original video and reconstructed video.Studies have shown that SSE or peak signal-to-noise ratio(PSNR)can measure data fidelity better.Existing research on human visual characteristics shows that the subjective quality of video is not completely consistent with the fidelity of the data.SSE as the distortion measure can measure data fidelity better,but it cannot fully guarantee better visual effects.Therefore,many researchers have proposed video coding methods based on visual perception quality.However,these methods only improve the visual perception quality of the video,but ignore the data fidelity.Because the structural similarity index(SSIM)is more consistent with human subjective.According to the local texture characteristics of video content,this paper proposes an RDO coding method based on the SSE and SSIM hybrid distortion measures,in order to improve the visual perception quality while ensuring the fidelity of the data.The experimental results show that under the configurations such as all intra(AI),random access(RA),and low delay(LD),compared with the testing software HM-16.14 of H.265/HEVC,the proposed method has better subjective visual quality.Compared with the method of only using SSIM distortion measure,the proposed method can effectively guarantee data fidelity.Specifically,under the same SSIM of reconstructed video,compared with HM-16.14,the coding rate of the proposed method decreases by 2.919%(AI),3.182%(RA)and 3.673%(LD)on average;compared with the method of only using SSIM distortion measure,the coding rate of the proposed method only increases by 2.502%(AI),0.788%(RA)and-2.507%(LD)on average.Using data fidelity PSNR to measure video quality,under the same PSNR of the reconstructed video,compared with HM-16.14,the coding rate of the proposed method only increases by 0.727%(AI),1.245%(RA)and 1.768%(LD)on average;compared with the method of only using SSIM distortion measure,the coding rate of the proposed method decreases by 10.245%(AI),36.108%(RA)and 52.056%(LD)on average.Furthermore,based on the proposed RDO coding algorithm with hybrid distortion measures,this paper also proposes a rate control algorithm based on the hybrid distortion measures.The experimental results show that under the coding configurations such as RA and LD,compared with the traditional R-? model rate control algorithm used in HM-16.14,the proposed method has better subjective quality.Specifically,under the same SSIM of reconstructed video,compared with the traditional R-? model rate control algorithm used in HM-16.14,the coding rate of the proposed method decreases by 3.148%(RA)and 3.594%(LD)on average.Using data fidelity PSNR to measure video quality,under the same PSNR of the reconstructed video,compared with the traditional R-A model rate control algorithm used in HM-16.14,the coding rate of the proposed method only increases by 1.237%(RA)and 1.821%(LD)on average.Compared with the rate control algorithm of the traditional R-A model adopted in HM-16.14,the proposed method has similar rate error.