The famous and elegant Patch Match algorithm [1], seen on the course, finds a dense set of correspondences between two images, by iteratively updating an initial random disparity chart. These disparities are then “triangulated” to obtain a cloud of points in 3D space. The goal of this project is to develop a variant of the Patch Match algorithm that directly computes the cloud of 3D points, from an initial assignment of random heights on a fixed geographic grid. The algorithm can be trivially adapted to more than two images (without need of data fusion), by simply updating the cost function that evaluates a candidate height.
Supervision
Gabriele Facciolo, Carlo de Franchis, Enric Meinhardt
Bibliographic References
[1] Barnes, C., Shechtman, E., Finkelstein, A., & Goldman, D. B. (2009). PatchMatch: A randomized correspondence algorithm for structural image editing. ACM Transactions on Graphics (ToG), 28(3), 24.
