Research On HEVC Rate Control Algorithm Based On Visual Attention

With the high development of HD video applications,the requirements of video compression are getting higher and higher.Compared with H.264/AVC,the High Efficiency Video Coding is the latest generation of video compression standard with higher compression efficiency and better compression quality.HEVC uses a hybrid coding framework with varieties of technologies to achieve the ultimate goal of video compression.Rate control is an important technology in video coding.How to allocate the target bit rate to the coding unit at all levels to ensure the final bitstream meet the channel bandwidth is the main problem to be solved by rate control algorithm.Through the study of the characteristics of the human observation system,we can see that humans have a certain degree of bias when watching images.Humans often pay more attentions to moving objects and texture complex areas.Based on the hybrid coding framework,the HEVC video compression standard does not consider the subjective attention characteristics of the human.In this paper,we focus on the rate control algorithm with the background of the High Efficiency Video Coding standard.To solve the problem in the rate control algorithm,we made some improvements considering the subjective attention characteristics of human.At the same time,according to the difference between subjective observation and current quality evaluations,we also made some improvements on the video compression quality evaluation.The main work of this paper can be summarized as the following three points:1.An improved rate control algorithm based on the combination of subjective attention region and space-time domain is proposed in this paper.Based on the structure of the current rate control algorithm,the subjective region detection result combined with the space-time domain is used as the allocation weight of the CTU layer to modify the target bit allocation.In the time domain,a motion detection algorithm based on optical flow and global motion estimation is proposed.In the spatial domain,a weighting method based on texture complexity is proposed.And then,the target bit allocation weight combined with the time domain and the spatial domain is centrally enhanced.Finally,we proposed a new rate control algorithm considering the subjective observation of humans.2.This paper presents a new method of compression quality evaluation based on human subjective attention.Based on the result of the subjective observation experiment,we improved the evaluation standard of video compression quality.The weight of pixels is increased on the basis of PSNR objective quality evaluation standard.According to the saliency map of the eye movement experiment,we set different weight values for the pixels of different focus in a frame,and the pixel distortion with weight is used as the final evaluation parameter.This paper proposes two improvements on rate control and quality evaluation considering the subjective attention characteristics of the humans.Compared with the compression model provided by HEVC,the rate control algorithm proposed in this paper improves the quality of subjective compression without affecting the realization of the target bit.In addition,the quality evaluation algorithm proposed in this paper can improve the performance of the proposed rate control algorithm more intuitively and effectively.