《多视点最优纹理重构》青少年文学教育丛书.pdfVIP

Optimal Texture Map Reconstruction from Multiple Views Lifeng Wang Sing Bing Kang Richard Szeliski Heung-Yeung Shum Microsoft China Microsoft Corp. Microsoft Corp. Microsoft China Beijing, P.R. China Redmond, WA, USA Redmond, WA, USA Beijing, P.R. China Abstract texture from multiple reference images [6]. In another, Mor- ris and Kanade [13] find the best triangulation for a given The recovery of 3D models from multiple reference images set of feature point correspondences across multiple images. involves not only the extraction of 3D shape, but also of The metric used is the reprojection error for a given hypothe- texture. Assuming that all surfaces are Lambertian, the re- sized triangulation. Generation of correct textures is critical sulting final texture is typically computed as a linear combi- for such techniques. nation of reference textures. This is, however, not the optimal There has been a significant amount of work done on means for reconstructing textures, since this does not model generating an image with a resolution higher than its indi- the anisotropy in the texture projection. Furthermore, the vidual sources. This can also be considered as recovering an spatial image sampling may be quite variable within a fore- optimal texture map from multiple (smaller resolution) tex- shortened surface. This also has important implications for ture maps seen at different views. Existing super-resolution computer vision techniques that involve analysis by synthe- approaches can be categorized as being interpolation-based sis and