Image matching method and device based on graph neural network fusion model

Abstract

The invention relates to an image matching method and device based on a graph neural network fusion model, and the method comprises the following steps: 1), carrying out the entity target detection of an image through employing an entity detection algorithm, and obtaining image blocks containing entities; and extracting visual features of the entity image blocks by using a network based on visual feature extraction; 2) forming all entity image blocks in the image into a graph form, and further extracting spatial relation features by using a graph neural network; 3) aggregating the visual features of the entities to obtain a visual feature aggregation vector; and fusing the visual feature aggregation vector and the spatial relationship feature by using a feature fusion network, after fusion, calculating the Euclidean distance of the feature vector to measure the similarity between the image pairs, and if the Euclidean distance of the feature vector of the image pairs is smaller than a threshold value, predicting the image pairs to be matched, otherwise, predicting the image pairs to be not matched. According to the method, the interference of irrelevant contents in the image on matching is effectively avoided, and higher matching precision is achieved.

Description

Technical field [0001] The present invention relates to the field of image matching, particularly an image matching method and apparatus based on the fusion model of the map nerve network. Background technique [0002] Image matching refers to a method of determining an image pair similarity and consistency by a correspondence relationship between an image content, a feature, a structure, and the like. Image matches have a wide range of applications in the computer vision, such as image-based navigation, location identification, and closed-loop detection in the SLAM. Image matching methods can be divided into two categories: the first category is based on method of manual design features, using image descriptors to represent partial features, such as SIFT, SURF, and ORB; and further use all descriptors in the image using the polymerization model Features are polymerized, such as the phrase model (Bag-of-Words, BOW), vector of local polymerization descriptive, VLAD, Fisher vector ...

AI Technical Summary

Problems solved by technology

[0004] The purpose of this application is to provide an image matching method based on a graph neural network fusion model to solve the problem of low image matching accuracy in complex scenes

Images