Edge-guided second-order total generalized variation for Gaussian noise removal from depth map.

Area of Science:

• Computer Vision
• Image Processing
• Computational Mathematics

Background:

• Single image denoising has seen advancements with total generalized variation models.
• Depth map quality is crucial for 3D reconstruction and scene understanding.
• Existing denoising methods often struggle with preserving fine details and edges in depth maps.

Purpose of the Study:

• To develop an accurate and effective denoising method specifically for depth maps.
• To improve Gaussian noise removal while maintaining edge integrity in depth data.
• To enhance the performance of total generalized variation models for depth map processing.

Main Methods:

• A weighted second-order total generalized variational model is proposed for Gaussian noise removal.
• An edge indicator function is integrated into the regularization term to guide gradient diffusion.
• The first-order primal-dual algorithm is employed to minimize the energy function.

Main Results:

• The method achieves high-quality denoising for depth maps corrupted with high-intensity noise.
• Significant edge preservation is demonstrated, maintaining structural details.
• Quantitative and qualitative evaluations show superior accuracy and visual improvements over state-of-the-art methods.

Conclusions:

• The proposed weighted second-order total generalized variational model offers a robust solution for depth map denoising.
• The integration of an edge indicator function effectively enhances gradient diffusion and edge preservation.
• The method represents a significant advancement in depth map processing for computer vision applications.