Robust image matching constrained by delaunay triangulation
San Jiang, Wanshou Jiang · DOAJ (DOAJ: Directory of Open Access Journals) · 2020
Image matching is an important issue in the fields of photogrammetry and computer vision. This study exploits the usage of Delaunay triangulation for reliable image matching. First, randomly located initial matches are organized by using Delaunay triangulation, and neighboring connection relationships are established evenly and stably. Second, local photometric and geometric constraints are constructed based on the line descriptor and spatial angular order, which converts the problem of removing outliers to that of analyzing the similarity of the Delaunay triangulation and its corresponding graph. Third, a match expansion operation is implemented based on the local geometric constraint deduced from two corresponding triangles. Finally, a reliable image matching method is proposed with the assistant of the hierarchical elimination and cross-checking strategies. The proposed algorithm is verified by using three datasets, and the results demonstrate that even with high outlier ratios the proposed method can reliably remove false matches and provide match results with high precision.