Reconstructing textured meshes from multiple range+rgb maps
Marco Callieri, Paolo Cignoni, Roberto Scopigno · 2002
The reconstruction of high-quality 3D models sampling both the shape and the surface reflection attribute is still a complex task. This papers contains new algorithms to integrate in a single texturemapped mesh the color info contained in either standard digital photos or RGB-enhanced range maps, minimizing redundancy and optimizing the color attribute representation. We enhance a previous solution [12] with: improved use of the available images, better color matching and computational efficiency. According to visibility computations and some heuristics, we subdivide the mesh in patches and assign to each patch a section of one of the input images. These image sections are packed in a texture image minimizing space but maintaining image coherence. Finally, we eliminate most of the color difference or discontinuity which exist in input images (applying a cross-correlation and interpolation between adjoining sections of the input images, directly on the texture image space). Note that this last step is performed on the whole texture, in contrast to others approaches that apply color smoothing just on small transition zones. 1