Method for automatically eliminating wrongly-matched registration points in remote sensing image

Abstract

The invention provides a method for automatically eliminating wrongly-matched registration points in remote sensing image. The method is characterized by comprising the following steps of: obtaining images; extracting feature points, and building the initial matching relationship; searching K nearest feature points from a set U by an Euclidean space distance method; judging whether a point pair {PixA<0>, PixB<0>} and n feature points around thereof conform to the angle self-consistent requirements or not; and obtaining a common point set U<right> and wrong common points U<wrong> and other steps. After matched point pair sets and feature point description vectors are obtained by performing any feature point extraction and a matching algorithm on two remote sensing images, wrongly-matched feature point pairs can be eliminated by using the method.

Description

technical field [0001] The invention belongs to the field of remote sensing image processing, and in particular relates to a method for eliminating false matching of automatic registration points of remote sensing images. Background technique [0002] In the multi-source remote sensing images, the geographic locations of the same-named features are not the same in most cases. There are two main reasons for this phenomenon: First, there are significant differences in the position and attitude of the sensor at the moment the sensor acquires the image. ; Second, due to the large difference in topographical fluctuations in the study area. Due to the existence of these two factors, there are certain difficulties in the registration of multi-source remote sensing images. At present, the main solution is to first use the DEM data of the research area to perform orthorectification on the remote sensing images, then select the same-named feature points in the remote sensing images w...

AI Technical Summary

Problems solved by technology

However, for remote sensing images with high spatial resolution, it is difficult to obtain DEM data with similar resolution, so there is a large error in its orthorectification, which causes great difficulties in the extraction and matching of the same-named points in subsequent images.

[0003] At present, the technology of automatically extracting the same-name point and object points of multi-source remote sensing images is developing rapidly, and a variety of same-point object point extraction algorithms have been proposed. At present, the more effective method is to carry out mobile calculation with a certain size window after the basic registration of the images. However, for high-spatial-resolution remote-sensing images in areas with large terrain fluctuations, the distance between objects with the same name often exceeds several hundred pixels. If the search range is too large, the geometric multiple of time-course is calculated. grows, eventually resulting in an incalculable

[0004] Another type of method is the feature point extraction and matching algorithm, among which SIFT (Scale Invariant Feature Transform) and SURF (Speed ​​up Robust Features) feature point extraction algorithms are the most widely used, in the image of 2000╳2000 these two algorithms can extract Thousands of feature points are obtained, but because there are a large number of similar features in the remote sensing image, such as vegetation, houses, etc., it is easy to have a large number of wrong matching point pairs in the subsequent feature point matching process, so it is necessary to correct the wrong matching point pairs For elimination, the more effective method at present is the RANSAC (Random Sample Consensus) method, which can effectively solve the elimination of mismatching points in areas with small terrain relief, but in areas with large terrain relief, due to the inability to establish accurate mathematical model, so this method is difficult to play a role, so this patent proposes a more effective algorithm

Images