High Accuracy Depth Acquisition And Application Research

Compared with the traditional 2D and 3D TV,multi-view video allows users freely to choice view in a certain range of angle,view the content of interest,so it can give users a 3D immersion visual impact of more powerful and rich stereo feeling and unusual feeling.Compared to the current HD video,multi-view video data volume will increase with the increase in the number of points of view,the current storage media and network bandwidth obviously can not meet the requirements of multi-view stereo video.Depth image based rendering(DIBR,Depth Image Based Rendering)technology is one of the key technologies of multi-view video,it effectively combines the texture images and the corresponding depth image information,so that each pixel in any new image can be obtained through the 3D-warping equation.With the advantages of saving data space and network bandwidth,high quality and high speed,the technology is considered to be a candidate for the next generation of three-dimensional video.However,obtaining high quality depth map is very challenging,and its accuracy directly affects the rendering quality of stereoscopic video.For the above problems,the main work and results of this paper are as follows:1.An image enhancement algorithm based on Kinect for high accuracy depth image acquisition is proposed,which uses morphological filtering and bilateral filtering to fill the hole of the background texture map from Kinect.Then the foreground and background of the depth image obtained from the Kinect were separated using the three-dimensional Gauss mixture model(3DGMM,3D Gauss mixture model).Making different treatment for the area in the hole.The background hole is filled with the corresponding background value,and the foreground hole is filled with the algorithm of color matching and structural similarity to obtain the reliable depth image.2.An improved method of virtual viewpoint synthesis based on directional texture is proposed,the texture map and depth map of the virtual viewpoint are obtained after the 3D mapping of the texture map and the final depth map are registered,the hole in new images are divided into foreground hole and background hole according to it’s existing area,the background region is filled with background values and an improved image inpainting technique based on directional texture is used to repair the foreground holes,which make it get better image.3.A technique for multi-view color correction based on region of interest is proposed and evaluated its quality in this paper.Firstly,we got the source image and target image of multi-view video.Then extracted the interest region’s saliency map of the target image and match it with the corresponding region of source image.At last,the subjective and objective quality evaluation of the matched image is carried out.The experimental results show that compared with the traditional 3D Gauss mixture model,the objective quality evaluation of the new proposed algorithm is in better agreement with the subjective perceptual quality.