Supervised discrete cross-modal hashing based on kernel discriminant analysis
作者：
Highlights：
• We propose a supervised discrete cross-modal hashing framework which can establish strong and effective connection between different modalities and preserve the discrete constraint, thus reducing the quantization loss.
• A compact optimization strategy is presented to directly learn the hash codes in a closed form, rather than bit by bit.
• The evaluation on four real-world datasets demonstrates the superior performance of SDCH-KDA over the state-of-the-arts methods. Especially on the LabelMe dataset, SDCH-KDA promotes an average of 9% improvement compared to the best results available.
摘要
•We propose a supervised discrete cross-modal hashing framework which can establish strong and effective connection between different modalities and preserve the discrete constraint, thus reducing the quantization loss.•A compact optimization strategy is presented to directly learn the hash codes in a closed form, rather than bit by bit.•The evaluation on four real-world datasets demonstrates the superior performance of SDCH-KDA over the state-of-the-arts methods. Especially on the LabelMe dataset, SDCH-KDA promotes an average of 9% improvement compared to the best results available.
论文关键词：Cross-modal hashing,Discrete,Kernel discriminant analysis
论文评审过程：Received 28 September 2018, Revised 27 June 2019, Accepted 22 September 2019, Available online 23 September 2019, Version of Record 27 September 2019.
论文官网地址：https://doi.org/10.1016/j.patcog.2019.107062
