MLRank: Multi-correlation Learning to Rank for image annotation

作者：

Highlights：

• We formulate image annotation into a Multi-correlation Learning to Rank framework.

• It is a semi-supervised learning method leveraging labeled and unlabeled images.

• The ranking function is estimated by leveraging labeled and unlabeled images.

• The ranking function explores correlation consistencies between images and tags.

• Experiments on three benchmarks demonstrate the superior performance of our work.

摘要

Highlights•We formulate image annotation into a Multi-correlation Learning to Rank framework.•It is a semi-supervised learning method leveraging labeled and unlabeled images.•The ranking function is estimated by leveraging labeled and unlabeled images.•The ranking function explores correlation consistencies between images and tags.•Experiments on three benchmarks demonstrate the superior performance of our work.

论文关键词：Image annotation,Learning to rank,Multi-correlation,Image-bias consistency,Tag-bias consistency

论文评审过程：Received 8 November 2011, Revised 31 January 2013, Accepted 22 March 2013, Available online 6 April 2013.

论文官网地址：https://doi.org/10.1016/j.patcog.2013.03.016