Similar image recognition method and device

Abstract

The disclosure provides a similar picture recognition method and device, belonging to the field of picture recognition. The method includes: extracting feature points from two pictures to be identified to obtain multiple feature points of the two pictures; obtaining multiple first feature point pairs of the two pictures according to the multiple feature points of the two pictures; determining multiple The second feature point pair; when the number ratio between multiple second feature point pairs and multiple first feature point pairs is greater than the first preset threshold, and the number of multiple second feature point pairs is less than the number threshold, according to The connection angles of multiple second feature point pairs are used to filter multiple second feature point pairs to obtain multiple third feature point pairs; when multiple third feature point pairs and multiple second feature point pairs When the quantity ratio of is greater than the second preset threshold, it is determined that the two pictures are similar pictures. The disclosure avoids the situation of high recognition error rate in the process of recognizing scenes with fewer feature points, and improves the speed and accuracy of similar picture recognition.

Description

technical field [0001] The present disclosure relates to the field of image recognition, and in particular, to a similar image recognition method and device. Background technique [0002] With the development of information technology, similar image search is a hot topic in the current search field. The premise of similar image search is the recognition of similar images. Recognition of similar images based on feature algorithms is gradually becoming the mainstream research direction of similar image recognition methods, such as SIFT (Scale Invariant Feature Transform, using scale-invariant features) or SURF (Speeded Up Robust Features, accelerated robust features), etc. Taking the recognition of similar pictures based on the SIFT algorithm as an example, the method includes: extracting the SIFT point of each picture; using the Euclidean distance of the SIFT point as the similarity determination measure of the two pictures. For example, take a certain SIFT point Dot11 in pi...

AI Technical Summary

Problems solved by technology

However, when there are few feature point pairs, such as when there are only a few or a dozen pairs of feature points, in the process of screening the feature point pairs using the RANSAC algorithm, the samples randomly selected from the feature point pair set have It may be a wrongly matched feature point pair, and the probability of this error is relatively high. Therefore, the error rate of the feature point pair screened by the RANSAC algorithm increases, resulting in a decrease in the accuracy of similar image recognition.

Images