With the rapid growth of multimodal data, the cross-modal search has widely attracted research interests. Due to its efficiency on storage and computing, hashing-based methods are broadly used for large scale cross-modal retrieval. Most existing hashing methods are designed based on binary supervision, which transforms complex relationships of multi-label data into simple similar or dissimilar. However, few methods have explored the rich semantic information implicit in multi-label data to improve the accuracy of searching results. In this paper, the multi-level semantic supervision generating approach is proposed by exploring the label relevance. And a deep hashing framework is designed for multi-label image-text cross retrieval tasks. It can simultaneously capture the binary similarity and the complex multi-level semantic structure of data in different forms. Moreover, the effects of three different convolutional neural networks, CNN-F, VGG-16, and ResNet-50, on the retrieval results are compared. The experimental results on an open source cross-modal dataset show that our approach outperforms several state-of-the-art hashing methods, and the retrieval result on the CNN-F network is better than VGG-16 and ResNet-50. INDEX TERMS Cross-modal retrieval, deep learning, hashing method, multi-label learning.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.