Graph construction by incorporating local and global affinity graphs for saliency detection

作者：

Highlights：

• We proposed a saliency method by modeling three-stage graphs with the same nodes.

• First-graph integrates deep-level graph and low-level graph and optimized with learning graph.

• Second-graph embeds the saliency cues enhanced with a global affinity graph.

• Third-graph initially designed with cross-fusing deep-level graph and low-level graph.

• Third-graph considers the saliency cues enhanced with a global affinity graph.

摘要

•We proposed a saliency method by modeling three-stage graphs with the same nodes.•First-graph integrates deep-level graph and low-level graph and optimized with learning graph.•Second-graph embeds the saliency cues enhanced with a global affinity graph.•Third-graph initially designed with cross-fusing deep-level graph and low-level graph.•Third-graph considers the saliency cues enhanced with a global affinity graph.

论文关键词：Saliency detection,Graph construction,Multi-view features,Joint global affinity matrix,Local affinity graph

论文评审过程：Received 12 August 2021, Revised 20 February 2022, Accepted 8 April 2022, Available online 22 April 2022, Version of Record 28 April 2022.

论文官网地址：https://doi.org/10.1016/j.image.2022.116712