Vision for Graphics

Creating compelling-looking content using conventional graphics techniques is often laborious and requires significant artistry and experience. Two related issues that we’ve been investigating are how this content-creation process can be simplified without sacrificing quality, and how rendering can be made photorealistic or compelling. In this paper, we describe how images and videos can be processed using computer vision techniques to produce a variety of outputs, such as higher dynamic range video, realisticlooking interpolated views of dynamic scenes, 3D models of plants and trees, and rain. We also describe how flash/noflash image pairs can be easily used to produce mattes of foreground objects and how photos of paintings can be analyzed to facilitate animation.