On-the-fly feature importance mining for person re-identification

作者：

Highlights：

• The first study systematically investigates feature importance for person re-identification.

• An unsupervised approach for on-the-fly mining of attribute-specific feature importance is proposed.

• Computing selective feature weighting on-the-fly for each probe can improve re-identification.

• Existing generic feature weighting approaches and our method can play a complementary role.

摘要

Highlights•The first study systematically investigates feature importance for person re-identification.•An unsupervised approach for on-the-fly mining of attribute-specific feature importance is proposed.•Computing selective feature weighting on-the-fly for each probe can improve re-identification.•Existing generic feature weighting approaches and our method can play a complementary role.

论文关键词：Person re-identification,Unsupervised salience learning,Feature importance,Random forest

论文评审过程：Received 6 April 2013, Revised 4 October 2013, Accepted 1 November 2013, Available online 13 November 2013.

论文官网地址：https://doi.org/10.1016/j.patcog.2013.11.001