Perceptual Video Coding Algorithm Based On SRP-JND Model

The main goal of video compression is to efficiently condense visual data as much as possible while minimizing the loss of visual quality due to compression. Conventional video coding techniques mainly remove spatial redundancy, temporal redundancy and statistical redundancy. This kind of coding techniques improves the coding performance by increasing the computational complexity of the codec, such as High Efficiency Video Coding (HEVC). However, the improvement of the performance is becoming slower and slower. In recent years, many researchers have begun to try incorporating the perceived characteristics of human visual system (HVS) into video coding framework and formed a new field of research: perceptual video coding.After years of development, the research of perceptual video coding can be classified into the following two groups. The first group includes visual attention models based approaches such as saliency or foveation video coding. The researchers try to find out saliency target or object through saliency detection, form a saliency map and use the saliency map to guide bit allocation of video coding, which could reduce the bit allocation for non-saliency regions. The second group includes Just Noticeable Distortion (JND) based approaches that utilize redundancy information such as perceptually insignificant component from the pictures. Human eyes are usually not sensitive to distortion below the JND threshold. According to this theory, more bits should be allocated to the areas where coding distortions can be easily detected whereas fewer bits to the areas where coding distortions are less noticeable.In previous researches, almost all the works mentioned above either focus on visual sensitivity model or visual attention model for video coding only. A few studies combine both visual sensitivity model and visual attention model together. To address this problem, this study proposes a novel Saliency Region Priority Coding Based JND Model (SRP-JND) that combines visual sensitivity model and visual attention model to achieve a high compression rate while maintaining the subjective video quality. The saliency target or object will be found out by saliency detection and the JND threshold will be calculated by JND model, and then the saliency is utilized to adjust the JND threshold. For saliency region, the threshold of JND will decrease, while the threshold of JND for non-saliency region increases.A perceptual video coding based on SRP-JND has been proposed by incorporating the SRP-JND model into video coding framework. In framework of proposed perceptual video coding, each prediction residual is adjusted by a SRP-JND adaptive filter. In order to verify the effectiveness of the proposed perceptual video coding, this study implements it in HEVC test model HM14.0. With respect to the original HM14.0, the proposed perceptual video coding obtains an average of12.25%the bitrate saved while maintaining nearly the same subjective video quality. Experimental results show that the proposed perceptual video coding based on SRP-JND model can reduce the video coding bitrate and improve the video coding efficiency while maintaining the subjective video quality.