Image local feature matching method, device and terminal based on non-geometric constraints

Abstract

A method, device, and terminal for matching local features of an image based on non-geometric constraints, the method comprising: performing local feature detection on a target image and an image to be matched to obtain respective local feature points; The respective local feature points are subjected to local feature matching to obtain a plurality of first candidate matching feature points and a plurality of second candidate matching feature points, wherein each first candidate matching point is matched with a corresponding second candidate matching feature point, and the first A candidate matching feature point is located in the target image, and a second candidate matching feature point is located in the image to be matched; performing neighbor constraint detection on a plurality of first candidate matching feature points and a plurality of second candidate matching feature points will meet the neighbor constraint. The first candidate matching feature point and the second candidate matching feature point matching it are added to the matching feature point set. The technical scheme of the invention effectively realizes the mismatch elimination in various types of images.

Description

technical field [0001] The present invention relates to the field of image processing, in particular to an image local feature matching method, device and terminal based on non-geometric constraints. Background technique [0002] In the field of computer vision, local features of an image refer to the features of the local area of ​​the image; local usually refers to the pixel area that can appear stably and has good distinguishability. The local feature matching method has been widely used in many problems in the field of computer vision, and has become a necessary means in image retrieval, target detection and recognition and other application fields. [0003] In the prior art, local feature matching is divided into two stages: feature vector matching and mismatch elimination. Feature vector matching refers to the process of calculating feature vectors of feature points through descriptors, and then finding matching feature points according to the similarity between featu...

AI Technical Summary

Problems solved by technology

Since the non-rigid target can be deformed, the affine transformation relationship of the non-rigid target between the two images cannot be determined at this time, and it is difficult to eliminate the mismatch through geometric constraints, resulting in low target detection accuracy

In addition, when the relationship between two images is not a simple affine transformation, such as when the image contains repeated targets and multiple rigid targets, the RANSAC algorithm cannot eliminate the mismatch, which reduces the target detection rate

Images