Faithful Spatio-Temporal disocclusion filling using local optimization

We present a novel method to fill disoccluded regions occurring in Depth Image Based Rendering (DIBR) in a faithful way. Given a video stream and a corresponding depth map, DIBR can render arbitrary new views of a scene. Areas that are not visible in the reference view need to be filled after warping. We present a novel framework for this task which can reconstruct the disoc-cluded regions by taking temporally neighboring frames into account. An efficient optimization scheme is employed to find faithful filling regions. This way, in contrast to common methods, we can fill disocclusions with their true color values, yielding high-quality view synthesis results.