Accurate salient object detection via dense recurrent connections and residual-based hierarchical feature integration

作者：

Highlights：

• A novel dense recurrent CNN module (D-RCNN) is developed to construct more informative saliency cues.

• A residual-based architecture with short connections is presented to hierarchically integrating multi-level feature maps.

• The end-to-end method can produce accurate saliency maps without relying on any pre/post-processing techniques.

• More accurate salient object detection results are achieved with significantly fewer model parameters.

摘要

•A novel dense recurrent CNN module (D-RCNN) is developed to construct more informative saliency cues.•A residual-based architecture with short connections is presented to hierarchically integrating multi-level feature maps.•The end-to-end method can produce accurate saliency maps without relying on any pre/post-processing techniques.•More accurate salient object detection results are achieved with significantly fewer model parameters.

论文关键词：Convolutional neural network,Recurrent convolutional layer,Salient object detection,Hierarchical feature fusion,Deep supervision

论文评审过程：Received 18 November 2018, Revised 28 March 2019, Accepted 12 June 2019, Available online 14 June 2019, Version of Record 27 June 2019.

论文官网地址：https://doi.org/10.1016/j.image.2019.06.004