Dual-Level Semantic Transfer Deep Hashing for Efficient Social Image Retrieval

Social network stores and disseminates a tremendous amount of user shared images. Deep hashing is an efficient indexing technique to support large-scale social image retrieval, due its deep representation capability, fast retrieval speed low storage cost. Particularly, unsupervised has well scalability as it does not require any manually labelled data for training. However, owing the lacking label guidance, existing methods suffer from severe semantic shortage when optimizing large neural parameters. Differently, in this paper, we propose Dual-level Semantic Transfer Hashing (DSTDH) method alleviate problem with unified hash learning framework. Our model targets at semantically enhanced codes by specially exploiting user-generated tags associated Specifically, design complementary dual-level transfer mechanism efficiently discover potential semantics seamlessly them into binary codes. On one hand, instance-level are directly preserved adverse noise removing. Besides, image-concept hypergraph constructed indirectly transferring latent high-order correlations images Moreover, obtained simultaneously discrete optimization strategy. Extensive experiments on two public datasets validate superior performance our compared state-of-the-art methods. The source can be https://github.com/research2020-1/DSTDH