Image registration method and image registration device based on cascade structure

Abstract

The invention provides an image registration method and an image registration device based on a cascade structure. The method comprises the steps: carrying out bilateral matching processing on a characteristic point of a reference image and that of an input image to obtain a first coarse matching group; eliminating matched pairs in the first coarse matching group according to a predetermined confidence coefficient to obtain a second coarse matching group, wherein the similarity of the matched pairs eliminated from the first coarse matching group is lower than the predetermined confidence coefficient; eliminating the matched pairs in the second coarse matching group according to a predetermined threshold value to obtain a third coarse matching group, wherein the gray distance or the space distance between any two matched pairs in the matched pairs eliminated from the second coarse matching group is greater than the predetermined threshold value; processing the matched pairs in the third coarse matching group by adopting an RANSAC (RANdom SAmple Consensus) algorithm; and aligning the reference image with the input image spatially according to a transformational matrix so as to obtain an output image.

Description

Technical field [0001] The present invention relates to the field of image registration, in particular to an image registration method and device based on a cascade structure. Background technique [0002] Image registration is a basic problem of image processing, which is used to align two or more images acquired at different times, different sensors or under different conditions in space. Image registration has a wide range of practical significance and applications, and its applications involve remote sensing image processing, medical image processing, target recognition, panoramic image stitching and other fields. [0003] At present, the commonly used image registration algorithm is based on the feature point registration, based on the Scale-Invariant Feature Transform (SIFT) algorithm combined with the random sampling consensus (RANdom SAmple Consensus, RANSAC) algorithm is based on feature points The registration method, the realization process first uses the SIFT algorithm...

AI Technical Summary

Problems solved by technology

[0004] However, the RANSAC algorithm has requirements for the accuracy of rough matching. When the image is a multi-spectral image or the shooting angle of the image is quite different, and the accuracy of the rough matching obtained by the SIFT algorithm is very low, it cannot be obtained directly by using the RANSAC algorithm. Correct results, or even completely wrong results, resulting in spatially aligned reference images and the input image, the resulting output image resolution is not high, or even completely unclear

Images