Co-Salient Object Detection With Co-Representation Purification.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Image Analysis

Background:

• Co-salient object detection (Co-SOD) identifies common objects in related images.
• Accurate co-representation mining is crucial for Co-SOD.
• Existing methods often include irrelevant information in co-representations, hindering performance.

Purpose of the Study:

• To develop a method for obtaining noise-free co-representations in Co-SOD.
• To improve the accuracy and robustness of co-salient object detection.

Main Methods:

• Propose Co-Representation Purification (CoRP) to search for noise-free co-representations.
• Identify pixel-wise embeddings likely belonging to co-salient regions.
• Iteratively refine co-representations by reducing irrelevant embeddings using predictions.

Main Results:

• CoRP achieves state-of-the-art performance on benchmark datasets.
• Experimental results demonstrate the effectiveness of the purification method.
• The proposed approach enhances the precision of co-salient object localization.

Conclusions:

• The CoRP method successfully purifies co-representations for improved Co-SOD.
• This technique addresses the limitation of irrelevant information in current Co-SOD methods.
• CoRP offers a significant advancement in co-salient object detection accuracy.