| Image-based view synthesis is an emerging research topic, with numerous applications. The ability to synthesize new views from existing images, enables the generation of enhanced and synthetic environments. Approaches for view synthesis relay upon dense or sparse matching of the views. In both cases, some parts of the images are inevitably unmatched. Triangulation of sparsely matched points present a possible solution for the handling of those unmatched regions. However, such triangulation should respect the underlying geometry of the scene. In this work, a novel approach is proposed for physically valid triangulation of sparsely matched points. The proposed approach is based on the maximization of a physical validity criterion which is supported by textured regions in the images. The produced triangulation is such that each triangle corresponds approximately to a planar surface in the scene. Given an arbitrary initial triangulation, the proposed approach refines it by flipping the edges of triangles. Furthermore, since missing matched points may preclude the correct triangulation of the scene, an additional stage handles the addition of matched points inside low score triangles. Inherent to this approach, the support region which is used for the evaluation of the correctness of the added matches is normally much larger than the one used in a local match evaluation. A set of tests where conducted to evaluate the quality of the resulting triangulation under different conditions, such as in depth rotations of the observed scene and also, different kind of scenes and textures. Since the principal application of this method is image-based view synthesis, the results of the triangulation where demonstrated using different two kinds of methods, view interpolation and view extrapolation. The results of the method are visually satisfying for both view synthesis methods. |
