Helly hypergraph based matching framework using deterministic sampling techniques for spatially improved point feature based image matching

Hypergraphs are tools for matching of point-features incorporating spatial relationships in the form of hyperedges exhibiting topological and geometric features between the points of images to be matched. Considering all possible hyperedges is computationally expensive and are randomly chosen in the state of the art techniques. A Helly Hypergraph based Matching Framework (HHMF) is proposed for the matching of images using point-features with effective hyperedges. The framework includes proposed algorithms such as Construction of Hyperedges using Point-features by Random (CHPR), Combinatorial (CHPC), and Exhaustive (CHPE) sampling techniques with and without Helly selection. The resultant hyperedges are treated with Adaptive Block Co-ordinate Ascent Graph Matching with Integer Projected Fixed Point algorithm. The performance of the proposed framework is evaluated in terms of Accuracy, Matching score, Execution time and Tensor Size for synthetic point sets and Willow wine image dataset. Based on the experimental studies carried out against existing framework, CHPC, and CHPE with Helly selection, exhibited better performance with 73.88% & 81% accuracy for 53.64 & 14.8% reduced tensor size respectively, in deformation noise tests, and 98% & 96% accuracy for 97% & 70% reduced tensor size in outlier tests. In the implicit experimental comparisons within sampling techniques, CHPR, and CHPE provided better performance with 81.37%, and 76% accuracy. In general, HHMF framework has reduced the tensor size and execution time for deterministic sampling cases during point sets matching. The framework can be extended in the near future by incorporating learning schemes for automated hypergraph based point sets matching.