Hierarchical Feature Fusion Network for Salient Object Detection.

Area of Science:

• Computer Vision
• Deep Learning
• Image Processing

Background:

• Convolutional Neural Networks (CNNs) excel in salient object detection by leveraging high-level semantic information.
• However, pooling operations in CNNs can degrade feature map resolution, blurring object boundaries and hindering saliency map accuracy.
• Existing methods struggle to balance semantic richness with precise boundary localization.

Purpose of the Study:

• To propose a novel Hierarchical Feature Fusion Network (HFFNet) for salient object detection.
• To effectively fuse hierarchical features, retaining both semantic and edge information for improved saliency mapping.
• To address the limitations of pooling operations in CNNs for salient object detection.

Main Methods:

• The HFFNet extracts features from different levels of the VGG network.
• Features are fused hierarchically to generate both high-level semantic and low-level edge information.
• A one-to-one hierarchical supervision strategy is employed for feature generation, and edge information guides saliency map generation.

Main Results:

• The HFFNet demonstrates effectiveness in salient object detection across five benchmark datasets.
• Experimental results show superior performance compared to 10 state-of-the-art models.
• The edge guidance fusion effectively identifies saliency regions, improving map accuracy.

Conclusions:

• The proposed HFFNet effectively integrates semantic and edge information for salient object detection.
• The network achieves state-of-the-art performance, outperforming existing models.
• HFFNet offers a promising approach for accurate salient object detection by preserving fine-grained details.