Global refinement of image-to-geometry registration for color projection

The management, processing and visualization of color information is a critical subject in the context of the acquisition and visualization of real objects. Especially in the context of Cultural Heritage, artifacts are so complex or hard-to-handle that the appearance information has to be extracted from a set of images. The images usually have to be registered to the 3D model of the objects, in order to transfer the needed information. Hence, the problem of image-to-geometry registration has been thoroughly studied by the Computer Graphics and Computer Vision community. Several methods have been proposed, but a fully automatic and generic solution is still missing. Moreover, small misalignments often lead to visible artifacts in the final colored 3D models. In this paper, we propose a method to refine the alignment of a group of images which has been already registered to a 3D model. Taking advantage of the overlapping among the images, and applying a statistical global method based on Mutual Information, the registration error is distributed among all the elements of the dataset. Hence, the quality of color projection is improved, especially when dealing with small details. The method was tested on a number of heterogeneous Cultural Heritage objects, bringing to a visible improvement in the rendering quality. The method is fully automatic, and it does not need powerful hardware or long processing time. Hence, it represents a valid solution for a wide application on CH artifacts.