A Quick and Automatic Image Based Modeling and Rendering System

Abstract

We present a complete and automatic system for object capture from video, Internet delivery, and real-time photorealistic rendering. Shape from silhouette is used to capture an approximate enveloping geometry, the visual hull. Texture coordinates are generated automatically under a minimum distortion criterion. To account for the difference between the true shape of the object and the captured visual hull we use Dynamic texturing, a viewdependent texturing method, which explicitly captures geometric deviations and applies a texture based correction. These models are then efficiently coded, and delivered over the Internet.