Image feature matching method and device which combine SURF and ORB

Abstract

The invention relates to an image feature matching method and device which combine SURF and ORB. The method comprises the following steps of: inputting a first image and a second image; detecting feature points with scale invariance in the first image and the second image by SURF respectively; describing the feature points in the first image and the second image respectively by an ORB feature descriptor; completing an initial match according to a Hamming distance and a cosine similarity between a descriptor of the feature point in the first image and a descriptor of the feature point in the second image; eliminating mismatch in the initial match according to a PROSAC algorithm; generating a matching relationship of feature points in the first image and the second image. It can effectivelyimprove the accuracy and speed of image feature matching, while reducing the rate of mismatch.

Description

technical field [0001] The invention relates to the technical field of image detection, in particular to an image feature matching method and device combining SURF and ORB. Background technique [0002] In recent years, image matching has become a hot topic in the field of computer vision. Image matching technology is an important branch of image processing applications, and has a wide range of applications in many fields, such as image restoration, object tracking, image stitching, image fusion, 3D reconstruction, and pattern recognition. Therefore, the research on image matching has important theoretical significance and practical application value. [0003] The basic and key issue of computer vision is the accuracy and speed of image matching. Image matching methods are divided into two methods based on regions and based on features. The latter method has better performance than region-based methods in terms of noise, rotation, scaling, illumination changes, and graysca...

AI Technical Summary

Problems solved by technology

Vector distances can be shown to be similar, but it is difficult to describe the geometry of the spatial information

[0008] When eliminating mismatches, the RANSAC algorithm calculates the number of iterations of parameters without an upper limit, which increases the complexity of the calculation; if the upper limit of the number of iterations is set, the result may not be the best result, and may even get wrong results. It needs to be set Thresholds associated with the problem

Moreover, the original image matching algorithm based on binary features has the limitation of limited matching constraints. There are multiple similar regions in the local image, and the error matching rate is high. It is difficult for RANSAC to effectively remove error points.

Images