JND Model And Its Application In HEVC Perceptual Video Coding

With the rapid development of Internet and multimedia information technologies,people have higher demanding on video quality,which brings great challenges for the video coding technology.Therefore,the newest video coding standard termed high efficiency video coding(HEVC)was developed.Compared with its predecessor H.264/AVC,HEVC doubled the encoding efficiency.However,many new techniques in HEVC still exploit the statistical correlations of video signals to remove the spatial and temporal redundancy for achieving high coding efficiency.Considering that the human eyes are the ultimate receivers of video,the researchers have tried to explore the human visual system(HVS)characteristics to improve video coding efficiency.Among them,the just noticeable difference(JND)model can make more accurate quantization evaluation to visual redundancy in image/video.The JND model simulates main characteristics of HVS to obtain the minimum visual threshold of human eyes,and below which,human eyes are insensitive to the change.Thus,an effective JND model will be beneficial to estimate the visual redundancy accurately for higher video coding efficiency.Therefore,this thesis is to study more reliable JND model and its application in HEVC perceptual video coding,as follow:1.Most existing JND models in pixel domain overvalued the visual masking effect from orderly textural regions and ignored the visual attention characteristics of HVS,which thus failed to evaluate visual redundancy comprehensively.Hence,this thesis proposed visual attention guided pixel-wise JND model.Firstly,the original image was decomposed into structural image,orderly textural image and disorderly textural image to calculate corresponding spatial contrast to get the primary contrast masking model(CM Model)via relative total variation model(RTV Model)and pattern complexity.Secondly,given that saliency model is closely related to human eye's visual attention characteristic,the saliency adjust factor was used to tune contrast masking estimator adaptively.Finally,by combining the luminance adaptation(LA),the proposed JND model was established.The experimental results illustrated that our model estimated visual redundancy more accurately.2.Considering that the image blocks with different textural complexities represented by DCT block energy may lead to different degrees of the contrast masking,this thesis developed a novel DCT-based JND model via block energy.Firstly,based on the traditional block classification scheme,the types of the textural blocks were further divided according to block energy,and the extent of contrast masking effect is then measured.Secondly,the contrast masking model was built with new block partition method.Finally,luminance adaptation and spatial contrast sensitivity function were incorporated into our JND model.Experimental results demonstrated that the proposed JND model was more reliable.3.Through in-depth analysis,it can be found that the joint consideration of different intra prediction modes in HEVC and pixel-wise JND model could estimate the visual redundancy more accurately and thus improve HEVC encoding efficiency more effectively.Therefore,this thesis presented an HEVC intra perceptual video coding method based on direction-based JND model.First,according to the relationship between the gradient values and the angles of angle prediction modes,the direction-based JND model was developed.Secondly,the degree of visual redundancy of the current largest coding unit(LCU)in HEVC encoder was judged by the proposed JND model.Finally,Lagrange multiplier adaptively adjusted by the simplified Weber's Law to guide the rate distortion optimization process based on its degree of visual redundancy.Experimental results showed that the proposed method can reduce encoding rate effectively,while keeping the perceptual video quality.In summary,the proposed JND models can estimate the visual redundancy more accurately,meanwhile,the application of JND model in HEVC also significantly improved perceptual video encoding efficiency.The works in this thesis is of a certain research significance and is beneficial to the development of perceptual image/video coding technologies.