Depth-Quality-Aware Salient Object Detection.

Area of Science:

• Computer Vision
• Artificial Intelligence

Background:

• Existing RGB-D salient object detection methods use bistream structures for fusion.
• These methods are often unaware of depth data quality, leading to poor performance with low-quality depth information.

Purpose of the Study:

• To develop a depth-quality-aware subnet for selective RGB-D fusion.
• To improve salient object detection by addressing the limitations of depth quality variations.

Main Methods:

• Integrated a novel depth-quality-aware subnet into a bistream architecture.
• Assessed depth quality prior to selective RGB-D fusion.

Main Results:

• The proposed method effectively lessens the impact of low-quality or unhelpful depth regions during fusion.
• Achieved a significantly improved complementary status between RGB and depth data.

Conclusions:

• The depth-quality-aware approach enhances RGB-D salient object detection performance, especially in challenging scenarios with varying depth quality.
• This method offers a more robust and accurate solution compared to state-of-the-art bistream techniques.