Bidirectional neighborhood filtering policy based remote sensing image registration method

Abstract

The invention discloses a bidirectional neighborhood filtering policy based remote sensing image registration method. The bidirectional neighborhood filtering policy based remote sensing image registration method comprises the steps of: combining grey characteristics with spatial structure characteristics of characteristic points; iterating to obtain a matching point pair with the same bidirectional neighborhood structure through the difference of the bidirectional neighborhood structure by regarding a grey characteristics based matching result as an initial matching point pair and regarding neighborhood structural characteristics of the characteristic points as limitations; and recovering candidate wrong matching point pairs which are deleted wrongly in the iteration process by adopting a bidirectional neighborhood filtering policy. The bidirectional neighborhood filtering policy based remote sensing image registration method is used under the conditions that relatively great affine transformation exists in the to-be-registered image, the to-be-registered image is a heterogenous image and similar patterns exist in an image scene, and can increase registration precision under the condition of no manual intervention.

Description

technical field [0001] The invention belongs to the field of image processing, and relates to an image registration method, in particular to a remote sensing image registration method based on a two-way neighborhood filtering strategy. Background technique [0002] As an important preprocessing technique in image analysis, image registration is widely used in the fields of image fusion, computer vision and object recognition. The steps of image registration mainly include: feature extraction, feature matching, transformation model parameter estimation, image resampling and inverse transformation. Among them, the purpose of feature matching is to ensure a reliable mapping relationship between the image features to be registered, which is a key step in image registration. [0003] Due to the particularity of remote sensing images, there are many interference factors in the registration process of remote sensing images. First of all, remote sensing images come from a variety ...

AI Technical Summary

Problems solved by technology

Most of the current feature extraction methods are not stable when there is a large affine transformation between the images to be registered

In addition, due to the irrelevant grayscale of heterogeneous images and similar patterns in the scene, the accuracy of the matching method based on grayscale features is reduced

The registration method based on the spatial structure mainly uses the characteristics of the spatial structure for feature matching, such as the RANSAC algorithm, but because the RANSAN algorithm has strong constraints when judging the misalignment points, many misalignment points cannot be eliminated, thereby reducing registration accuracy

[0005] In summary, in view of the problems of large affine transformation, heterogeneous images, and similar patterns in the scene in remote sensing images, it is difficult for existing image registration methods to achieve high-precision registration.

Images