Fast multi-view image rendering method based on reverse search for matching

Abstract

In a traditional multi-view image generation algorithm, partial image information might be lost at the pixel mapping step during the 3D image acquisition. A lower hardware cost and shorter operation time can be realized if an effective algorithm is employed. Therefore, a fast multi-view image rendering method based on reverse search is proposed in this paper, it uses pixel mapping information to derive a rendering image. Compared with the conventional depth-image-based rendering (DIBR) which using multiple images, the method only generates one image with reverse mapping in the image rendering step. First, the parallax range of the original image is calculated. Then, a warping image is generated through reverse mapping; this image has contained information for eight different viewpoints. Finally, the image modification can be made. Experimental results show that the proposed method effectively reduces rendering time and memory size by 58.82% and 83.78%, respectively. In addition, the mean structural similarity between the two images created by DIBR and our method is 1, which means that our method has the same effect as DIBR.