Latent multi-feature co-regression for visual recognition by discriminatively leveraging multi-source models

作者：

Highlights：

• We propose a novel multi-feature representation co-regression framework by exploiting multi-source knowledge discriminately.

• We uncover multiple optimal latent spaces and minimizes the co-regression residual by considering correlations among multiple feature representations.

• The proposed method can automatically select the discriminative sources for each target feature domain via formulating a row-sparsity pursuit problem.

• A generalization error bound is also derived for our extension.

摘要

•We propose a novel multi-feature representation co-regression framework by exploiting multi-source knowledge discriminately.•We uncover multiple optimal latent spaces and minimizes the co-regression residual by considering correlations among multiple feature representations.•The proposed method can automatically select the discriminative sources for each target feature domain via formulating a row-sparsity pursuit problem.•A generalization error bound is also derived for our extension.

论文关键词：Multi-source adaptation,Multi-feature representation,Latent space,Group sparsity

论文评审过程：Received 14 February 2018, Revised 2 September 2018, Accepted 21 October 2018, Available online 25 October 2018, Version of Record 31 October 2018.

论文官网地址：https://doi.org/10.1016/j.patcog.2018.10.023