Research On Visual Saliency Extraction And Its Application In Video Coding

Traditional video compression methods remove spatial and temporal redundancy based on the signal objective statistical correlation.What's more,the traditional rate control technique treats each bit of the image in an equal manner,and there is no difference between the bits resources.However,the human watching the video image is affected by the human visual system mechanisms,and the visual attention is not assigned to each region of the image,but it is concentrated in some areas with high salient value.The main content of this paper is the extraction of visual saliency and its application in video coding.In this paper,we propose a new saliency model based on Curvelet Transform,which can accurately detect the salient regions in the natural scene.Then,a saliency model for the video is proposed,which is used to adjust the quantization parameters of the macroblock layer,which improves the subjective quality of the encoded video.The main work of this paper includes the following three aspects:(1)Propose a novel saliency detecting model based on Curvelet Transform: Firstly,the multi-scale curvelet coefficients are obtained by using the Curvelet Transform,and then the inverse transform is used to reconstruct the multi-scale feature map.Finally,the final saliency map is obtained by the corresponding mathematical calculation.(2)Propose a novel method to adjust the quantization parameters of macroblock layer based on saliency: Firstly,a saliency model for video is proposed,which includes spatial saliency extraction,temporal saliency extraction and the fusion method between them.Finally the saliency value of each macroblock is obtained by the fusion between the spatial and temporal saliency map,then the saliency is used to adjust the quantization parameters of the macroblock layer.(3)In the experiment,the weighted EWSSIM based on eye tracking is introduced as the subjective performance index.Large numbers of experiments show that: the saliency model proposed can accurately extract the salient region in video image,and the video encoding based on saliency can maintain the objective quality and improve the subjective quality of video.