A comprehensive evaluation index generation method for image registration

Abstract

The invention discloses a comprehensive evaluation index generation method for image registration. The method comprises the following steps of: carrying out random selection in initial matching feature points to obtain a subset, and calculating a transformation matrix by utilizing the subset; carrying out a matching error of each matching feature point pair and a mean value of all the matching errors, and obtaining an accumulative error elimination index Z through the amount of former smaller than later; calculating the sum of distances between different matching feature points in a reference image, dividing the reference image into image blocks, and carrying out statistic on differences between proportions of the matching feature points and a maximum value and a minimum value, so as to obtain a distribution index P; calculating a matching quantification error of each matching feature point pair and summing the matching quantification errors so as to obtain a matching quantification error index O; calculating a mean value of matching errors of all the matching feature points and solving a mean value quantification error index R; and combining the indexes Z, P, O and R to calculate a final comprehensive evaluation index RE. According to the method, the problem that RMSE similar evaluation indexes are influenced by the amount of feature points and the mean values of errors.

Description

technical field [0001] The invention belongs to the technical field of image processing, and relates to an evaluation index for image registration, in particular to a method for generating a comprehensive evaluation index for image registration, which can eliminate the "cumulative error" of feature point matching evaluation, measure the distribution of matching feature point pairs, and measure The matching feature point pairs with different errors have different effects on the registration results, and the measurement results of the matching feature point pairs can be matched with the measurement results of the image registration results. Background technique [0002] Because the same sensor shoots a certain target at different times, or different sensors shoot a certain target, there will be differences in positions, angles, etc., and even image distortion often occurs. This difference and distortion will greatly interfere with image change detection, image splicing and fus...

AI Technical Summary

Problems solved by technology

At the same time, the existing evaluation indicators are generally single-evaluation matching feature point pairs (such as single threshold method, root mean square error, etc.), or only evaluate the final registration result (such as mutual information), without matching feature points Evaluation method corresponding to the final registration effect

Images