Image retrieval method, system and device

Abstract

The application discloses an image retrieval method, system and device. The method includes: carrying out classification judgment on a to-be-retrieved image through an overall retrieval model; selecting a corresponding local retrieval model for the to-be-retrieved image according to a classification judgment result of the overall model; carrying out feature analysis on the to-be-retrieved image bythe corresponding local retrieval model, and carrying out hash code mapping to obtain a hash code of the to-be-retrieved image; at the same time, retrieving hash codes, which are similar to the hashcode of the to-be-retrieved image, in a target hash code library corresponding to the local retrieval model; and finally, selecting a similar hash code set, which satisfies a similar condition, therefrom, and finding a similar image set, which corresponds to the similar hash code set, in an image database to complete retrieval. According to the method, the overall retrieval model of classificationjudgment is pre-established, then the overall retrieval model is utilized to generate the local retrieval model, retrieval accuracy for the image is improved, the target hash code library corresponding to the local retrieval model is established, and the retrieval accuracy is further improved.

Description

technical field [0001] The invention relates to the field of image retrieval, in particular to an image retrieval method, system and device. Background technique [0002] With the advent of the era of big data, Internet image resources have grown rapidly, and the concentration and scale of image data resources have brought opportunities and challenges to image retrieval. Compared with the traditional text-based image retrieval management and query, it is very difficult and difficult to meet the actual requirements. Therefore, Content-Based Image Retrieval (CBIR) has been extensively studied by many scholars. A series of issues such as how to effectively describe the feature information of images, and which data structure to use for efficient indexing and fast similarity retrieval have become research hotspots in this direction. [0003] In the prior art, CNNH and CNNH+, a supervised hashing method, are proposed, which first factorize the semantic similarity matrix of pairs...

AI Technical Summary

Problems solved by technology

[0003] In the prior art, CNNH and CNNH+, a supervised hashing method, are proposed, which first factorize the semantic similarity matrix of pairs of training image data into approximate hash codes, and then use these approximate hash codes and image labels Training a deep convolutional network, but the matrix decomposition in CNNH and CNNH+ will bring additional errors, making the training target deviate; and another hash method directly based on CNN, DLBHC, uses a threshold to binarize a fully connected layer as a binary result, resulting in a hash code; however, the above method still performs poorly on indistinguishable images

Images