Projective transformation image matching method based on transformation-invariant low-rank texture

Abstract

The invention discloses a projection transform image matching method based on a transform invariant low-rank texture, and mainly solves the defects in the prior art that projection transform image matching cannot be completed. The scheme is that 1. two images containing projection transform are inputted and automatic detection and extraction of low-rank texture areas are respectively performed; 2. TILT transform is performed on the low-rank texture areas obtained through detection so that respective local transform matrixes are obtained and the inputted two images are corrected by using the local transform matrixes; 3. feature point detection is performed on the two corrected images, and scale-invariant feature descriptors and geometric shape descriptors are established for feature points; and 4. new feature descriptors are established through combination of the scale-invariant feature descriptors and the geometric shape descriptors, and similarity measurement is performed on the new descriptors by using Euclidean distance so that image matching is completed. The feature points with relatively high repetitive rate and correct matching rate can be extracted so that computational efficiency can be enhanced, and the method can be applied to image fusion, image splicing and three-dimensional reconstruction.

Description

technical field [0001] The invention belongs to the technical field of image processing, and in particular relates to a projective transformation image matching method, which can be applied to the fields of target recognition and tracking, image splicing and three-dimensional reconstruction. Background technique [0002] In the fields of target recognition, image stitching and 3D reconstruction, it is necessary to match multiple views of the same scene. In general, feature-based image matching methods can be used for image matching, mainly because some image features are invariant to image scale, rotation and affine transformation, and only using feature information to find the geometric relationship between images has The advantages of high computational efficiency. However, when there is a large degree of projection distortion between the two images, it is often difficult to extract features with projection invariance in the existing technology, resulting in insufficient ...

AI Technical Summary

Problems solved by technology

[0007] The purpose of the present invention is to address the shortcomings of the above-mentioned prior art that cannot complete projection transformation image matching, and propose a projection transformation image matching method based on transformation invariant low-rank texture, which eliminates projection distortion of the input image through TILT transformation, and projects The matching problem of transformed image is transformed into the image matching problem of similar transformation to obtain more accurate matching point pairs

Images