Image retrieval method and device based on self-coding pre-dimension reduction and storage medium

Abstract

The invention discloses an image retrieval method and device based on self-coding pre-dimension reduction and a storage medium. The method comprises the steps of extracting the image features of the original image by using a pre-trained feature extraction network; sending the image features to a self-coding network before iterative quantization is carried out on the image features; and through secondary learning of the self-coding network, realizing the pre-dimension reduction of image features to extract key features, converting the key features into the binary hash codes through iterative quantization, and then outputting an image retrieval result according to a hamming distance between the binary hash codes and the binary codes of the reference image. Through the self-coding network, the pre-dimension reduction is realized, the input dimension of the iterative quantization is reduced, the loss of the feature information during the iterative quantization is avoided, and meanwhile, due to the fact that the input dimension of the iterative quantization is less, the time consumed by calculation is reduced, so that the retrieval accuracy and the retrieval efficiency are greatly improved.

Description

technical field [0001] The invention relates to the field of image recognition, in particular to an image retrieval method, device and storage medium based on self-encoding pre-dimension reduction. Background technique [0002] At present, with the development of image recognition technology, content-based image retrieval is one of the methods that can retrieve images from massive databases. Since this retrieval method does not need to manually annotate images, it has high intelligence. It is widely used in military, architectural design and face recognition systems. The traditional method mainly uses the computer to extract the underlying visual features of the image for recognition, but most of these underlying visual features are global features. When it comes to more complex retrieval requirements, deeper image features need to be extracted, but common deep features For example, LBP and HOG cannot associate pixel-level information with semantic information perceived by ...

AI Technical Summary

Problems solved by technology

The traditional method mainly uses the computer to extract the underlying visual features of the image for recognition, but most of these underlying visual features are global features. When it comes to more complex retrieval requirements, deeper image features need to be extracted, but common deep features For example, LBP and HOG cannot associate pixel-level information with semantic information perceived by humans, that is, there is a problem of semantic gap.

[0003]In order to solve this problem, the convolutional neural network is usually used to extract features in the prior art. Although this solution can solve the problem of semantic gap, the convolutional neural network The extracted features are usually high-dimensional, and directly used for image retrieval will easily lead to the loss of more feature information when the image is quantized and encoded, which will affect the accuracy of image retrieval.

Images