Shape from silhouette under varying lighting and multi-viewpoints
We present a new segmentation method for visual hull (VH) reconstruction from an image sequence. The camera moves on a hemisphere around an object set on a turntable with complex and unknown background. Even though the background is assumed to be constant, it cannot be modeled from all possible viewpoints. A series of images of a moving object under different lighting conditions are captured and analyzed. In this analysis, information from space, time and light domains are fused into an MRF framework. The method is validated by VH reconstruction of challenging objects with different optical properties.