Passive reconstruction of high quality textured 3D models of works of art

Abstract

A wide-spread use of 3D models in cultural heritage application requires low cost equipment and simple modeling procedures. In this context, passive 3D reconstruction methods allow to build 3D models from a set of calibrated cameras, without the need of expensive machinery. Unfortunately the surfaces characteristics often lead to bad quality reconstructions. Recent efforts attempt to combine together information from different passive methods in order to improve the overall quality of the result. The combination of stereo matching and silhouette information has recently received considerable attention. Typically the major contribution to the appearance of the model comes from texture, rather than from geometry. The straightforward application of the photographs over the model can lead to artifacts, due to errors in 3D reconstructions, which must be minimized. This work, building on recent results, proposes a variation of an algorithm for 3D geometry recovery from stereo and silhouette information within a classical deformable model framework, which improves the quality of the shape. In order to avoid visible texture artifacts, it also proposes a new algorithm for texture synthesis based on wavelet decomposition. Experimental verification shows the effectiveness of the proposed solution with respect to robustness, computational speed and quality of the final result.