Stereoscopic Image Generation Based On Depth Maps

Compared to many traditional 3D systems, in which left and right images are all needed to be transmitted, DIBR system only requires to transport left (or right) image and the associated depth maps, and right image (or left) image can be induced from them. In DIBR system, depth image is a 2D image that gives depth value of each point on an object in real scene. As this kind of image can be coded more efficiently than left or right image itself, DIBR method can reduce the bandwidth requirement compared with traditional 3D image system. This makes it not only suitable for 3D-TV, but also for other 3D applications such as Virtual reality, military, medical image processing, video effects.DIBR method can generate right images (left images) through 3D-Warp. However, there is a problem that occlusion holes appear after pixel to pixel mapping. Average filter is commonly used for hole-filling. But when there are numbers of big size holes nearby the edges, this approach will make the edges blur. Another novel method uses Gaussian filter to smooth the whole depth image before 3D image warping. This work can reduce the number of big holes in maps. However, many of experimental results show that if we smooth the whole depth map before warping, the final warped image quality is bad. For example, it will lead to geometric distortion at vertical edges. Edge filtering method based on depth image can improve the image quality, but the premise is accurate edge detection in the horizontal direction.In this paper, we analyzes the fundamental reasons that the changing of depth value from small to large or large to small in the horizontal direction lead to holes in warped images. Then, a weighted iteration method is proposed to deal with these locations. This new method can not only avoid geometric distortion caused by Gaussian filter, but also be realized easily because complicated edge detection is unnecessary.Experimental results demonstrate that the proposed method gets high quality in warped image and the PSNR is better than the previous work. Therefore, stereoscopic generation based on weighted iteration method of depth images can promote the application of DIBR system.