Abstract

There are emerging interests from both computer vision and computer graphics communities in obtaining photorealistic modeling of a scene or an object from real images. This paper presents a tentative review of the computer vision techniques used in such modeling which guarantee the generated views to be geometrically correct. The topics covered include mosaicking for building environment maps, CAD-like modeling for building 3D geometric models together with texture maps extracted from real images, image-based rendering for synthesizing new views from uncalibrated images, and techniques for modeling the appearance variation of a scene or an object under different illumination conditions. Major issues and difficulties are addressed.