DEEP HASHING MULTI-LABEL IMAGE RETRIEVAL WITH ATTENTION MECHANISM

Multi-label image retrieval is one of the important approaches in the field of image data processing to find similar images from image databases according to a given query example. However, the large amount of data and high-latitude features cause a large amount of calculation, and the extracted image...

Full description

Saved in:
Bibliographic Details
Published in:International journal of robotics & automation 2022-07, Vol.37 (4)
Main Authors: Xie, Wu, Cui, Mengyin, Liu, Manyi, Wang, Peilei, Qiang, Baohua
Format: Article
Language:English
Subjects:
Data processing
Feature extraction
Image management
Image retrieval
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Multi-label image retrieval is one of the important approaches in the field of image data processing to find similar images from image databases according to a given query example. However, the large amount of data and high-latitude features cause a large amount of calculation, and the extracted image feature vectors contain complex background information, resulting in low accuracy of multi-label image retrieval. To solve this problem, the attention mechanism is introduced to present a novel method of deep hashing multi-label image retrieval in this paper. Firstly, the residual network model is constructed with deep hash using the attention mechanism to identify the approximate positions of multiple objects from multi-label images. Secondly, the loss functions of class cross-entropy are utilized to improve the deep learning models via attention mechanisms. Experimental results show that the proposed method has higher accuracy on two multi-label data sets than other benchmark models, which plays important roles in the image data processing fields.
ISSN:0826-8185
1925-7090
DOI:10.2316/J.2022.206-0589