Keypoint Matching for Non-Rigid Object via Locally Consistent Visual Pattern Mining
| Autor: | Jun Shimamura, Tetsuya Kinebuchi, Shin'ichi Satah, Shuhei Tarashima |
|---|---|
| Rok vydání: | 2018 |
| Předmět: |
Matching (graph theory)
Computer science business.industry Node (networking) ComputingMethodologies_IMAGEPROCESSINGANDCOMPUTERVISION Local property Pattern recognition 0102 computer and information sciences 02 engineering and technology Object (computer science) 01 natural sciences Graph Visualization 010201 computation theory & mathematics 0202 electrical engineering electronic engineering information engineering Graph (abstract data type) 020201 artificial intelligence & image processing Artificial intelligence Pattern matching business Cluster analysis |
| Zdroj: | ICIP |
| DOI: | 10.1109/icip.2018.8451016 |
| Popis: | Keypoint matching is challenging especially when objects are non-rigid because underlying deformation is unknown and its complexity is varying. Finding visually similar (less deformed) regions between an image pair is an intuitive and promising approach, but this observation has not been fully exploited in existing methods. In this work, we propose a novel keypoint matching method for non-rigid objects via discovering such visual patterns shared in the image pair. We solve this problem by recursively finding clusters over a graph where each node represents a tentative match and each edge encodes geometric consistency between matches. Our method exploits the local property of target patterns to efficiently build the graph, and extends a local clustering algorithm to take 1-to-l keypoint matching constraints into account. Experimental results on public datasets show the flexibility and the efficiency of our approach to various deformation patterns. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|