Multi-source Image Matching Based On Structural Constraints

Multi-source images are widely used in remote sensing,military and medical.Multisource image matching is one of the most important fundamental problems in cross-field image processing.In this scenario,the specific characteristics of the same scene for different source images do not guarantee the similarity.Therefore,directly using general image feature matching methods for multi-source image matching may result in a lot of mismatches.In our work,in the two-dimensional structural space and the high-dimensional structural space separately,we propose structurally refined feature matching schemes for multi-source images,which improve matching accuracy.For multi-source image matching,most of the feature based matching methods just consider local feature comparison between images but neglect the structural correlations of feature points within one image.On the other hand,if we only use structural similarity for matching,exhaustive enumeration of all possible matching pairs is needed,which results in high computational complexity.To address these deficiencies,we combine the advantages of feature and structure matching,and present the refined structure matching framework.The main contents are as follows:(1)Based on coarse feature matching,we develop a two-dimensional structural constraint for refining the multi-source feature matching.Our structural refinement rejects the small number of mismatches which contradict the spatial consistency.Experimental results validate that our method not only properly refines the feature correspondences between multisource images,but also outperforms alternative state-of-the-art cross-field matching methods.(2)We develop a hypergraph matching framework which effectively refines different scales of multi-source image features rough matches.Two-dimensional structural constraints cannot accurately depict the space of similarity at different scales.In order to further enhance the effectiveness of structural constraints,we establish an association hypergraph based on the feature point correspondences.We then reject mismatches by identifying outlier vertices of the hypergraph through higher order clustering.For matching-different scaled images,our method is invariant to scale variation of objects and computationally more efficient than existing hypergraph matching methods.In summary,our structural constrained matching framework,to a certain extent,improves the accuracy in two-dimensional structure space with the same scale and in high-dimensional structure spatial with different scales of the multi-source image matching.