Belief Propagation on the GPU for Stereo Vision and
Image-Based Rendering
Abstract: The potential of the application of Markov Random Field (MRF) formulations to low-level vision problems, such as stereo, has been know for some time. However, recent advances, both algorithmic and in processing power, have made their application practical. This paper presents a novel implementation of Bayesian Belief Propagation (BP) for graphics processing units (GPU) found in most modern desktop and notebook computers, and applies it to the stereo problem and to Image-Based Rendering (IBR) for panoramic image generation and view interpolation of complex scenes. The stereo problem is used to compare results for accuracy and running time to other BP algorithms.