Maximum mean discrepancy regularized sparse reconstruction for robust salient regions detection

作者：

Highlights：

• We exploited visual and contextual information to precisely extract the salient objects.

• Our extracted background dictionary is very effective in removing background noises.

• We included a Laplacian term to preserve similarity and locality among salient regions.

• Our MMD regularized term transfers sparse coding to an effective representation.

摘要

Highlights•We exploited visual and contextual information to precisely extract the salient objects.•Our extracted background dictionary is very effective in removing background noises.•We included a Laplacian term to preserve similarity and locality among salient regions.•Our MMD regularized term transfers sparse coding to an effective representation.

论文关键词：Maximum mean discrepancy,Laplacian graph regularized,Sparse reconstruction error,Salient regions detection,Human visual attention

论文评审过程：Received 4 October 2016, Revised 28 February 2017, Accepted 28 February 2017, Available online 7 March 2017, Version of Record 10 March 2017.

论文官网地址：https://doi.org/10.1016/j.image.2017.02.013