视觉显著区域的图像编码.pdfVIP

ABSTRACT With the development of the times, video data and image data exist everywhere. Huge amounts of data have attracted people’s attention. The big data brings a lot of information and promotes the development rapidly. However, it also causes a lot of problems, such as storage problem and bandwidth problem. Therefore, image coding is still one of the main research topics. Common image coding method is aimed at the whole picture for compressing each object to the best. But in fact, the human eyes focus on a partial region of the image rather than all of them. That is called visual saliency. In this paper, we use the principle of visual saliency to make sure the quality of interest region at the cost of reduction of the quality in non-interest region. We proposed three kinds of implement method. First, the coding method based on quantization which controlling by depth. Directly use of depth map to be visual saliency map. That is to say, make the region closed to the image device as the region of interest and the other region as the region of non-interest. And distribute more bits to the region of interest through adjusting the quantization parameter values in the HEVC. The other region is distributed less. Second, the coding method based on the depth of field reconstruction. The method used HC method and depth information to get visual saliency map. That is region of interest map. And then reconstruct the region of field by the saliency map and depth map. Third, the coding method based on the principle of rate distortion optimization visual significant. Count the QP, rate bits and distortion in each block by encoding HEVC. Then use the saliency map as the weight to calculate the best QP by the rate distortion optimization theo