Image retrieval method and device

Abstract

The invention discloses an image retrieval method and device. The method comprises the steps: carrying out first feature extraction processing of a to-be-retrieved image and a database image through employing a trained feature extraction network, carrying out the regularization global attention sampling processing, carrying out second feature extraction processing through employing the trained feature extraction network, carrying out the feature dimension reduction processing to obtain a to-be-retrieved dimensionality reduction feature map and a database dimensionality reduction feature map, and performing similarity calculation on the two to obtain a retrieval result. The feature extraction network training process is as follows: performing scaling processing on the training image, and performing feature extraction on the scaled training image by using a feature extraction network to obtain a feature matrix; performing regularization attention sampling processing on the feature matrix to obtain a global structure sampling graph and a local structure sampling graph; and performing knowledge distillation processing based on the global structure sampling graph and the local structure sampling graph to obtain a trained feature extraction network. According to the method, scene local features can be deeply learned, and the retrieval accuracy is improved.

Description

technical field [0001] The invention relates to the technical field of image retrieval, in particular to an image retrieval method and device. Background technique [0002] This section is intended to provide a background or context to embodiments of the invention that are recited in the claims. The descriptions herein are not admitted to be prior art by inclusion in this section. [0003] In fields such as medical image review and financial review, a large number of pictures will be involved. The traditional method is to use the naked eye to check the relevant personnel. When the number of pictures is large, it may even be reviewed by random inspection. This method is very inefficient and incomplete, especially as the number of pictures involved reaches millions to tens of millions, it is time-consuming and labor-intensive to review manually, and it is impossible to effectively analyze such a huge number of pictures. Therefore, people propose to use the method of image re...

AI Technical Summary

Problems solved by technology

This global feature contains too much spatial information. There are many pictures in the process of medical image review and financial audit review, and the main feature network between scenes is not targeted. For example, there are repeated similar images, occlusions in the same scene, or Similar parts are blocked in a large area, seriously affecting the retrieval results, and the retrieval accuracy is not high

Images