Structural Consistency Based Aerial Remote Sensing Image Point Feature Matching

There are local pseudo-isomorphism,high outlier ratio,local repetitive structure,local structure distortion in the putative correspondences from the images with gray changes,low overlapping areas,repetitive patterns,viewpoint changes and local distortion.These problems make accurate point matching challenging.To improve the accuracy,efficiency and robustness of feature matching algorithm,three feature matching methods based on structural consistency are proposed as follows.To improve the accuracy and efficiency of feature matching with existence of local pseudo-isomorphism caused by gray changes and repetitive patterns,a discriminative point matching algorithm named local structure consensus constraint is proposed to remove the outliers from putative correspondences and find two local structure consensus graphs composed of inliers.Evaluated with 45 aerial image pairs with repetitive local patterns and ambiguous features,the proposed algorithm is proved to be more accurate and efficient.High outlier ratio feature points often lead to many mismatches in images with local repetitive patterns,similar target texture and low overlapping areas.To eliminate the mismatches accurately and efficiently,the relative motion between the putative matches and their K nearest neighbors are formulated.Then the relative motion entropy is defined to find the coincident relative motions.Based on relative motions with minimum relative motion entropy,the outliers with random or irregular relative motion are removed efficiently and accurately with quasi-linear time complexity.Compared with ten state-of-the-art feature matching algorithms,our MRME is proved to be more robust and accurate.To match the points with local repetitive structure from images with repetitive texture and repetitive patterns,and local distortion from the images with the wave and current.Based on the location changes of correctly matching feature points in different images,the relative motion and absolute motion between the putative matches and their K nearest neighbors are used to evaluate the global and local motion consensus.Tested with three typical datasets,the matching results indicate that the proposed global and local motion consensus feature matching method is more robust with the existence of heavy outliers.