Multi-View Stereo Using Discontinuity-Aware Planar Parameters

Hanshin Lim, Hyon-Gon Choo

Abstract: One of the successful approaches in the traditional field of Multi-View Stereo is PatchMatch-based methods. However, because the approaches rely heavily on photometric consistencies, the results are unreliable in low-textured areas. This paper proposes a Multi-View Stereo method that enforces the reliability of similarity costs by utilizing planar parameters refined through spatial consistencies in the PatchMatch process. In the proposed method, initially, depth maps and normals are generated by applying a conventional PatchMatch method. Afterward, a triangular mesh is generated from the selected nodes, and planar parameters are computed from the resulting triangular mesh model. The discontinuity-aware PatchMatch is then processed by leveraging the planar parameters refined through adjacent triangular meshes for similarity cost computation. The results of the experiments demonstrate that the proposed method achieves superior performance compared to existing Multi-View Stereo methods for publicly available High-Resolution images.