Research On Upsampling Algorithm For Depth Image Guided By Local Edge Preservation

With three dimensional television(3DTV)and three dimensional(3D)movies more and more widely entering human life,high-quality visual experience can provide consumers with sufficient spiritual support,so people have higher and higher requirements for the visual quality of images and videos.3DTV system needs to input two dimensional(2D)color video and 2D depth data from the same scene at the same time.Among them,the depth data can explain the location of the scene,and the stereoscopic display technology can provide consumers with a 3D stereoscopic vision.Therefore,depth data is an important basis for 3DTV systems,and the acquisition of high-quality depth information has received widespread attention.Depth information can be obtained by direct and indirect methods.Among them,due to the limitations of the hardware devices used to capture depth information in the direct method,such as the inability to effectively suppress noise interference and the high price,it cannot satisfy consumers to directly obtain depth information demand.The existing depth upsampling algorithms can achieve good results,but there are still problems such as noise interference,inefficient use of the correlation between pixels,and failure to achieve a detailed comparison of the depth map and color map in the edge area structure.In view of the current problems,this paper focuses on the analysis of the edge characteristics of the depth map,and obtains the high resolution(HR)depth image on the basis of maintaining the local edge guidance.The effectiveness of the proposed algorithms are verifiedin subjective and objective quality evaluation.The research works of this paper are as follows:1.Algorithm for depth image upsampling based on adaptive parameter guidance and edge point assistanceIn order to solve the problems such as the blurring of the edge doping in the depth edge region of the current algorithm,the mismatch between the structure of the depth image and the color image,and the generation of artifacts.In this paper,the edges of low resolution(LR)depth maps are used to complete the labeling and correction of unreliable pixels in LR depth images and pixel clustering,and to initialize the LR depth map.The initialization depth map is improved according to the region characteristics of the pixels,wherein the depth prediction is performed on the pixels in the complex structure region by using an autoregressive model with multiple adaptive parameters added.This algorithm can effectively reduce the ambiguity of the depth discontinuity region,and obtain stable and reliable high-quality HR depth images.2.Depth image upsampling algorithm based on depth edge points and color image guidanceFor the existing depth upsampling algorithms,the correlation between pixels is not considered,and the consistency criteria of the depth map and the color map are not accurate enough.This paper proposes a depth upsampling algorithm based on depth edge points and color map guidance.Firstly,the LR depth image is corrected for unreliable pixels,and the LR initialized depth image is obtained.Secondly,based on the gradient matrix of the HR color image,the structure consistency judgment of the depth map and the color map is achieved based on a33? pixel block centered on a certain pixel,and the pixels classification are completed.Finally,the spatial position constraint is used to set the influencing factor for the effective depth to complete the depth correction,and output the HR depth image with clear structure,reduced dead point rate,and effective enhancement of the detail edge region.