Abstract
Finding a correspondence between a pair of feature sets is an important problem with several applications ranging from camera calibration to panoramic image generation. When this problem is formulated as a transportation (or, minimum-cost network flow) problem, the optimal solution induces the sets of feature correspondences. Considering the case where multiple optimal solutions exist for a given transportation problem, each optimal solution yields different feature correspondences. The objective of this paper is to utilize deep learning to select the optimal solution, which best preserves local neighborhood information in such cases. Instead of a hand-crafted optimal solution selection as done by the previous work [6], we propose a novel deep learning-based optimal solution selection. The experimental evaluation of the proposed work including a comparison with the previous framework demonstrates both the effectiveness and efficiency of the overall approach.