Semi-supervised image retrieval method and device based on disturbance consistency self-integration

Abstract

The invention discloses a semi-supervised image retrieval method and device based on disturbance consistency self-integration, and the method comprises the steps: inputting an image into a trained semi-supervised image feature extraction model, and obtaining the features of the image, wherein the semi-supervised image feature extraction model comprises a convolutional neural network, a Hash layer and a disturbance consistency self-integration module; converting the features of the image into discrete binary hash codes of the image; and performing retrieval according to the binary hash code to obtain an image retrieval result. According to the invention, by integrating the features of the same sample under different data enhancement conditions, the discrimination feature of each category can be found; the similarity between the Hash layer output of the unmarked data and the corresponding integrated features is maximized through the designed disturbance consistency loss function, and the generalization ability of the network is improved by fully utilizing the unmarked data; and a better retrieval effect can be obtained.

Description

technical field [0001] The invention belongs to the technical field of software, in particular to a semi-supervised image retrieval method and device based on disturbance consistency self-integration. Background technique [0002] With the explosive growth of image data on the Internet, massive image data and high-dimensional image features make image retrieval face great challenges. Due to the characteristics of low storage cost and fast retrieval speed, deep hashing method has become a research hotspot in recent years. [0003] Generally, the deep hashing method achieves fast retrieval by mapping high-dimensional real-valued image features into compact binary hash codes, and uses the semantic similarity of images to constrain the hash codes during the mapping process to ensure retrieval accuracy. In the big data environment, supervised hashing methods often rely on a large amount of labeled image data to obtain high retrieval accuracy, but when there is only a small amoun...

AI Technical Summary

Problems solved by technology

In the big data environment, supervised hashing methods often rely on a large amount of labeled image data to obtain high retrieval accuracy, but when there is only a small amount of labeled data, the performance of supervised hashing methods will be greatly reduced

Chinese patent application CN109800314A discloses a method for generating a hash code for image retrieval using a deep convolutional network, which adds a hash layer before the classification layer, and the output of the hash layer is binarized to obtain the hash code of the image code, but the application is to use a large amount of labeled data to train the hash model to obtain better retrieval performance, but in actual scenarios, labeling a large amount of data requires huge human and material resources

[0005] At present, the above semi-supervised hashing method uses the visual similarity between samples to represent the semantic similarity between samples, but the visual similarity does not reflect the real semantic similarity between samples. Two samples with similar visual information may come from two different categories

Therefore, using the wrong visual similarity to guide the learning of hash codes will cause the similarity of hash codes learned by two samples to be inconsistent with the real semantic similarity relationship.

Images