Deep convolutional hashing using pairwise multi-label supervision for large-scale visual search

作者：

Highlights：

•

• We propose a new deep convolutional hashing approach by leveraging multi-label images and exploring the label relevance.

• The deep method combines convolutional networks and hashing layer to transform images into hash codes.

• The pairwise multi-label supervision in regularized loss function utilizes label relevance to compute semantic similarity of images.

• The experiments of visual search on two multi-label datasets demonstrate the competitiveness of our proposed approach.

摘要

•We propose a new deep convolutional hashing approach by leveraging multi-label images and exploring the label relevance.•The deep method combines convolutional networks and hashing layer to transform images into hash codes.•The pairwise multi-label supervision in regularized loss function utilizes label relevance to compute semantic similarity of images.•The experiments of visual search on two multi-label datasets demonstrate the competitiveness of our proposed approach.

论文关键词：Learning based hashing,Deep learning,Convolutional neural networks,Pairwise multi-label supervision,Label relevance

论文评审过程：Received 6 January 2017, Revised 10 May 2017, Accepted 23 June 2017, Available online 1 July 2017, Version of Record 7 November 2017.

论文官网地址：https://doi.org/10.1016/j.image.2017.06.008