PhD Program Summer School 2006
Global texture registration with mutual information maximization
speaker | Ioan Cleju |
date | September 29, 2006 |
abstract |
A common framework towards the synthesis of realistic 3D models requires
registration of un-calibrated photographic images to the 3D model. Camera
calibration recovers the camera parameters for each photograph, allowing
further texture registration. Commonly, camera calibration algorithms use
2D-3D correspondences chosen by users. The problem can be automatized
considering 2D-3D alignment methods, such as maximization of the mutual
information between the 3D model and the 2D image. We propose an optimization framework for global texture registration. Given a 3D model and several un-calibrated photographic images of the 3D model, we consider the problem of joint camera calibration by maximizing the global mutual information. The global mutual information consists of mutual information between images and the 3D model, and mutual information between overlapping images. We discuss several aspects of the optimization procedure and present a tool for texture registration of point-based 3D models. |