Probabilistic saliency estimation

作者：

Highlights：

• We propose a novel probabilistic framework for salient object detection.

• The solution to the proposed saliency criteria is shown to have closed form.

• We show that proposed method enjoys interpretations of graph cut, diffusion maps and one-class classification.

• The proposed method performs favorably to unsupervised state-of-the-art.

摘要

•We propose a novel probabilistic framework for salient object detection.•The solution to the proposed saliency criteria is shown to have closed form.•We show that proposed method enjoys interpretations of graph cut, diffusion maps and one-class classification.•The proposed method performs favorably to unsupervised state-of-the-art.

论文关键词：Saliency,Salient object detection,Spectral graph cut,Diffusion maps,Probabilistic model,One-class classification

论文评审过程：Received 30 January 2017, Revised 20 August 2017, Accepted 12 September 2017, Available online 20 September 2017, Version of Record 29 September 2017.

论文官网地址：https://doi.org/10.1016/j.patcog.2017.09.023