Learning to Associate Image Features with CRF-Matching

摘要: This paper presents a supervised learning algorithm for image feature matching. The is based on Conditional Random Fields which provides mechanism globally reason about the associations. novelty of this work twofold: (i) use Delaunay triangulation as graph structure probabilistic network to association; (ii) combination local and joint features enforce consistency in theoretically sound statistical procedure. Experimental results show that our approach outperforms RANSAC challenging datasets consisting indoor outdoor images, with significant occlusion, blurring, rotational translational transformations.