False Match Detection Method Based on L1 Norm Global Geometric Consistency Test

Abstract

A false match detection method based on L1 norm global geometric consistency test: use scale invariant feature transformation and bag of words model to extract and match feature points in two images; calculate the square distance matrix of two images; use golden Split search method to solve: minλ>0||D1‑λ.D2||1; calculate the difference matrix E=||D1‑λ*.D2||1; calculate the sum of the elements of each row of the E matrix and sort from large to small , and calculate the quadratic difference of each row sum value after sorting, take the point that reaches the maximum quadratic difference value as the threshold value, and if the sum of all rows is higher than the threshold value, the feature point pair corresponding to the row is judged as a wrong match Yes; after removing the wrong matching pairs, the similarity between the images is calculated according to the real matching point pairs of the two images, and then the image retrieval results are sorted from large to small according to the similarity. The invention allows complex backgrounds, partial occlusions and various similar geometric transformations between similar images; only the coordinate information of feature points is used, which is very simple and efficient.

Description

technical field [0001] The invention belongs to the field of image retrieval, in particular to a method for detecting wrong matching points between images in the field of partially repeated image retrieval. Background technique [0002] In recent years, the duplicate image search technology, including many search engines including Tineye, Baidu Map and Google similar image search, has developed rapidly. Wide range of applications. In this technology, the detection of mismatched feature point pairs between images is a key step. How to use the geometric information between images to correctly filter the wrong matches so as to obtain more accurate retrieval results is the core of this technology. . [0003] Partially repeated images mainly refer to pictures taken from different angles of the same scene or pictures before and after processing by image processing software. Such images are somewhat different in tone, lighting, scale, rotation, and occlusion, which makes it diff...

AI Technical Summary

Problems solved by technology

J.Philbin et al. [4] proposed to apply the classic Random Sampling Consistency (RANSAC) algorithm to deal with the problem of mismatch detection under the perspective transformation model, but using RANSAC will lead to a large calculation time-consuming, so it is not suitable for large-scale search questions under

Another way of thinking is the geometric coding (GC) method proposed by Wengang Zhou et al. [5]. This method first encodes the mutual position information and rotation transformation information of feature points in each image, and then compares the The encoding difference of the feature points is used to detect the wrong matching points. Since the scale and main direction information of the feature points are used, this method is still time-consuming.

The feature of the global method is that the detection effect is better, and it can adapt to more complex geometric transformation models. The disadvantage is that it consumes too much time and is not conducive to the application background of large-scale retrieval.

Images