Design and Implementation of Multimodel Estimation Algorithm for Nonrigid Matching Images

An algorithm of multimatching model estimation is proposed for the problem of outliers removed from nonrigid matching images. First, considering the high probability of outlier ratio in the matched point set, the inlier ratio promotion based on the consensus of the distribution of neighbor inliers was applied. Second, to reduce the influence of the inlier distance error threshold on the inlier extraction, a multimodel estimation was applied using the inlier distance error marginalization. Finally, considering the probability of residual outliers in the extracted inlier set, residual outliers were removed based on the consensus of the direction of the matching point position change vector. In the experiments, the proposed method is compared with MAGSAC, NM-NET, P-NAPSAC, SC-RANSAC, Adalam, OANET, SuperGlue, PEARL, Multi-H, Multi-X, and CONSAC, etc. Results indicate over 30% reduction in the inlier distance error, 50% reduction in the outlier residual rate, 8% increase in the recall of inlier, 10% reduction in the running time, and 16% reduction in the misclassification rate of multiplanar estimation.