Seamless Texture Optimization for RGB-D Reconstruction.

Area of Science:

• Computer Vision
• Computer Graphics
• 3D Reconstruction

Background:

• High-fidelity texture restoration is crucial for 3D models in AR/VR, cultural heritage, and entertainment.
• Existing texture mapping methods suffer from blurring, ghosting, and visual seams due to geometric errors and camera pose drift.

Purpose of the Study:

• To propose a novel tri-directional similarity texture synthesis method for RGB-D 3D reconstruction.
• To eliminate texture inconsistency and generate visually realistic texture mapping results.

Main Methods:

• Incorporation of a color image texture detail layer for enhanced context.
• Optimal texture image selection per triangle face, weighting texture details to avoid high-frequency area partitioning.
• Camera pose optimization for texture image alignment with the 3D shape.
• Tri-directional similarity function for resynthesizing image context within texture chart boundaries.
• Global color harmonization to address inter-viewpoint color inconsistencies.

Main Results:

• The proposed method effectively overcomes texture tearing, blurring, and ghosting artifacts.
• Experimental results demonstrate superior performance compared to state-of-the-art texture mapping techniques.
• Achieved visually realistic texture mapping with reduced seams and improved fidelity.

Conclusions:

• The novel tri-directional similarity texture synthesis method significantly enhances texture mapping quality in 3D reconstruction.
• The approach provides a robust solution for restoring high-fidelity textures, addressing limitations of existing algorithms.