PSRNet: a phase-guided frequency-domain structure reconstruction network for RGB-T salient object detection

Area of Science:

• Computer Vision
• Image Processing
• Machine Learning

Background:

• Salient object detection using RGB-Thermal (RGB-T) data is challenging due to structural shifts and weak thermal boundaries.
• Existing methods struggle with cross-modal feature alignment and boundary detail preservation.

Purpose of the Study:

• To propose PSRNet, a phase-guided frequency-domain structure reconstruction network for robust RGB-T salient object detection.
• To improve structural preservation and boundary recovery in challenging RGB-T scenarios.

Main Methods:

• Decomposing RGB and thermal features into amplitude and latent-phase components in the Fourier domain.
• Aligning cross-modal structural cues via amplitude-weighted phase consistency.
• Reconstructing high-frequency boundary information using a bounded Gaussian high-pass gate and adaptive phase-modulated fusion.

Main Results:

• PSRNet achieved an S-measure of 0.903, MAE of 0.028, and F-measure of 0.794 on the VT1000 dataset.
• Boundary IoU reached 0.862, demonstrating superior boundary recovery.
• The network showed improved structural preservation under challenging RGB-T conditions.

Conclusions:

• PSRNet effectively addresses limitations in RGB-T salient object detection.
• The proposed frequency-domain approach enhances structural and boundary details.
• The method offers a promising solution for complex visual environments.