Supervised deep hashing with a joint deep network

作者：

Highlights：

• Deep features learning module, deep mapping module and binary codes learning module are integrated into one unified architecture.

• The interaction between deep feature representations can be considered by deep neural network RNN.

• The proposed method can effectively reduce the information loss and directly generate binary codes without relaxation.

• The semantic similarity and balancing property of hash codes are preserved by new objective function.

摘要

•Deep features learning module, deep mapping module and binary codes learning module are integrated into one unified architecture.•The interaction between deep feature representations can be considered by deep neural network RNN.•The proposed method can effectively reduce the information loss and directly generate binary codes without relaxation.•The semantic similarity and balancing property of hash codes are preserved by new objective function.

论文关键词：Image retrieval,Supervised hashing,CNN,RNN,Deep learning

论文评审过程：Received 16 May 2018, Revised 3 March 2019, Accepted 6 April 2020, Available online 7 May 2020, Version of Record 7 May 2020.

论文官网地址：https://doi.org/10.1016/j.patcog.2020.107368