Depth Map Super-Resolution Considering View Synthesis Quality.

Area of Science:

• Computer Vision
• Image Processing
• 3D Graphics

Background:

• High-quality depth maps are crucial for 3D applications like multi-view rendering and 3D reconstruction.
• Captured depth images often have lower resolution than color images, limiting their performance.
• Existing methods struggle with noise and occlusion in low-resolution (LR) depth maps.

Purpose of the Study:

• To develop a novel depth map super-resolution (SR) method that enhances depth map quality.
• To improve the performance of 3D applications by addressing the resolution gap between depth and color images.
• To create a robust SR method that accounts for noise and occlusion in depth data.

Main Methods:

• A credibility-based multi-view depth map fusion strategy to refine LR depth maps, considering view synthesis quality and inter-view correlation.
• A view synthesis quality-based trilateral depth-map up-sampling method incorporating depth smoothness, texture similarity, and view synthesis quality.
• Refinement of LR depth maps to mitigate noise and occlusion before up-sampling.

Main Results:

• The proposed method outperforms state-of-the-art depth SR techniques in generating super-resolved depth maps.
• Synthesized views using the enhanced depth maps show significant quality improvements.
• The method demonstrates robustness to noise, achieving promising results even under noisy conditions.

Conclusions:

• The novel depth SR method effectively improves depth map resolution and quality for 3D applications.
• Considering view synthesis quality during fusion and up-sampling is key to superior performance.
• The approach offers a robust solution for depth map enhancement, particularly in the presence of noise.