Saliency and correlation learning for co-salient object detection

Ying Tong

doi:10.1016/j.engappai.2025.112504

Saliency and correlation learning for co-salient object detection

2025-10-03

0

PRE

AI

DOI：10.1016/j.engappai.2025.112504

Original

Original request for help

Save

Abstract

En 中文

Co-Salient object detection aims to identify common salient objects across a given group of images. However, accurately locating co-salient objects remains challenging due to the complexity of capturing the correlation representation of each group of images. To tackle this problem, we propose a saliency and correlation learning method for co-salient object detection. This method employs a saliency learning network and a correlation learning network to generate precise co-saliency maps of a group of images. Within the saliency learning network, a saliency feature grafting module is designed to refine object edges and achieve accurate detection of salient objects. Furthermore, the correlation learning network incorporates two modules, which are designed for extracting saliency correlation representation and deriving consensus correlation representation within a group of images, respectively. Guided by prior information obtained from saliency learning of images, our method significantly improves performance in co-salient object detection through correlation representation learning. Extensive experiments on all the latest benchmarks demonstrate that our method outperforms 11 state-of-the-art models, achieving a new level of technical excellence, with an average Structural Similarity Measure score of 0.845.

Saliency and correlation learning for co-salient object detection

Abstract

Journal

Organization