Rendering-based free-view image synthesis using peripheral view and depth images

Unlike light-field image captured with camera based on micro lens array, light-field image captured with a camera matrix has a large disparity, so it is difficult to synthesize a new view of image. When the new view image is generated from the light-field image having the large disparity by using the conventional depth image based rendering, there is a problem that a hole occurs in the warped color image. To solve this problem, we propose rendering-based free-view image synthesis using adjacent view and depth images. In the proposed algorithm, the camera position of the floating-point level of target view image, the image of four adjacent view images and each depth image are used as inputs. Then, warp the input images to the target view by projection and the projected images are synthesized to generate the image of target view. Unlike conventional methods, even if the target view image camera position is a floating-point, target view image is created that has no holes. Therefore, the proposed algorithm enables image synthesis at free-view, free from the limitations of the conventional stereo image synthesis method. And this can be confirmed by the experimental results.