Depth-based image processing for 3d video rendering applications

Abstract

Continuous advances in video rendering platforms and image processing techniques have made possible innovative multimedia services that are changing the way visual media is consumed. An interesting application is free-viewpoint television (FTV), which allows the viewer to interactively select the scene viewing angle. To allow smooth navigation within the scene, virtual views must be rendered between the actual cameras. One solution for generating 3D scenes is through the capture of multi-view videos and their corresponding depth maps, with the latter supplying scene geometry information, which is essential for view synthesis. In this paper, we present a depth image-based rendering (DIBR) algorithm which utilizes improved image processing techniques to synthesize virtual views with rendering quality surpassing that of current state-of-the-art algorithms. On average, a peak signal-to-noise ratio (PSNR) gain of 2.5 dB and 1.5 dB for the Ballet and Breakdancers sequences, respectively, is achieved when compared to current solutions.