Automatic Registration Method of Remote Sensing Image

Abstract

The invention relates to an automatic registration method for remote sensing images. The first registration image and the reference image after the registration of the images to be registered are divided according to a regular grid of a predetermined size, and the divided sub-image blocks are centered. The point of the same name closest to the lower right corner of the sub-image block is used as the intersection point of the grid division to generate an irregular rectangular grid, which improves the uniformity of the point of the same name, improves the registration accuracy of the area where the point of the same name is sparsely distributed, and then improves the overall image. Registration accuracy; then expand the irregular rectangular grid to generate a first-level sub-image block pair and perform registration, and cut the extended part of the registered first-level sub-image block to generate the revised first registration image The sub-image blocks, which may produce deformed edge areas after registration, are cut out, and finally the sub-image blocks of the revised first registration image are spliced ​​to obtain the registered image, which overcomes the sub-image block ground objects caused by local deformation of the image The incomplete correspondence problem further improves the image registration accuracy.

Description

technical field [0001] The invention relates to the field of image processing, in particular to an automatic registration method for remote sensing images. Background technique [0002] Image registration is one of the most important links in image processing, information integration and analysis. It is also the basis and important step of image stitching, image fusion, and super-resolution reconstruction technologies. It is used in computer vision, satellite image processing, and medical diagnosis and treatment. There are a wide range of applications in the field. High-precision image registration can make subsequent processing more efficient and the processing results better. Taking satellite imagery as an example, image registration is to select the reference image and the image to be registered in two or more multi-source images of the same scene, refer to the geometric position relationship between the reference image and the object, and use the predetermined matching ...

AI Technical Summary

Problems solved by technology

[0004] Although the image to be registered is registered in the related art, because the distribution of feature points in the image to be registered and the reference image is not uniform, the selected point of the same name is not uniform, and it is located near the point of the same name when performing quadratic polynomial correction. The pixels with the same name will have a higher correction accuracy, and the farther the pixel is from the same name point, the lower the correction accuracy will be.

Therefore, the use of regular grids to divide sub-image block pairs will result in lower uniformity of homonymous points in the divided sub-image blocks, resulting in lower registration accuracy in areas where homonymous points are sparsely distributed, and reducing the registration accuracy of the overall image. Accuracy; and because the registration may produce deformation of the edge area of ​​the sub-image block, it will lead to incomplete correspondence of the sub-image block's ground features, further reducing the registration accuracy of the image

Images