LRNet: lightweight attention-oriented residual fusion network for light field salient object detection

Area of Science:

• Computer Vision
• Artificial Intelligence

Background:

• Light field imaging offers rich scene structure information beneficial for salient object detection.
• Applying this information to salient object detection remains a significant challenge.

Purpose of the Study:

• To propose a novel network for salient object detection utilizing light field imaging.
• To enhance the accuracy and effectiveness of salient object detection in complex scenarios.

Main Methods:

• A lightweight attention-based feature enhancement module (LFM) generates attention maps for focal slices.
• A residual convLSTM-based feature integration module (RFM) fuses spatial-structural information from focal slices.
• The network combines LFM and RFM for comprehensive feature processing.

Main Results:

• The proposed method significantly enhances saliency features through focused attention.
• High-precision feature fusion is achieved by leveraging residual mechanisms and convLSTM.
• Experimental results demonstrate superior performance over 17 state-of-the-art methods on three public light field datasets.

Conclusions:

• The developed lightweight attention and residual convLSTM network effectively addresses challenges in light field salient object detection.
• The method achieves state-of-the-art performance, indicated by the highest scores across five quantitative metrics.