A Video-Based 3D Rendering Algorithm Using Hybrid Light-field and Geometry Rendering

This paper proposes a video-based rendering algorithm using hybrid light-field and geometry rendering.It synthesizes the first frame of a new view with a light-field rendering algorithm, and computes the foreground region by using the Gaussian Mixture Model and geometry computing.Updating the foreground region in the last synthesized frame results in a frame of the new view.This scheme effectively avoids re-rendering the background region which occupies a large area but changes slowly, and hence improves the rendering performance.To eliminate the accumulated error, a new cyclic pattern of "the first frame and the subsequent frames" is employed.In the meantime, the distribution of the foreground points of the scene is gained in a cyclic pattern, and is used to divide the scene of the next cycle.Experimental results show that our approach can synthesize high-quality images efficiently.