High-Fidelity Image-Based Modeling

This article presents a novel method for acquiring high-fidelity solid models of complex 3D shapes from multiple calibrated photographs. The proposed approach enforces both the photometric and geometric constraints associated with available image data using a simple iterative deformation process. Concretely, a widebaseline stereo matching technique based on Mikolajczyk’s and Schmid’s affine regions is first used to reconstruct a dense set of patches on the surface of the object of interest. Next, the boundary of the object’s visual hull is deformed to pass through the centers of the reconstructed patches and recover the surface’s main structural features and concavities. Fine surface details are finally reconstructed using a local refinement process that enforces smoothness, photometric, and geometric constraints at every vertex of the surface. The proposed approach has been implemented, and tested on 10 real datasets including objects with fine details, high-curvature areas, and deep concavities, and an object with little texture. Qualitative and quantitative comparisons with models obtained by stateof-the-art image-based modeling algorithms and laser range scanners are also presented.