Adaptive image warping for hole prevention in 3D view synthesis

Area of Science:

• Computer Vision
• Computer Graphics
• Image Processing

Background:

• 3D video popularization necessitates efficient monocular video conversion to stereoscopic or multi-view formats.
• Depth Image-Based Rendering (DIBR) is common but creates holes due to unknown 3D scene structure.
• Existing methods use regular grids for depth map deformation to hide holes, but this can be complex.

Purpose of the Study:

• To propose an adaptive image warping approach for hole-free 2D to 3D video conversion.
• To reduce the computational complexity of depth map deformation for virtual view synthesis.
• To maintain or improve the visual quality of generated 3D views.

Main Methods:

• Developed an adaptive image warping algorithm leveraging depth map smoothness.
• Utilized quadtrees to partition depth maps into homogeneous blocks.
• Applied an optimization process on an adaptive grid derived from these blocks.

Main Results:

• Significantly reduced computational complexity compared to regular grid deformation methods.
• Successfully prevented visible holes in synthesized 3D views.
• Maintained high visual quality in the resulting stereoscopic or multi-view video.

Conclusions:

• The adaptive image warping approach offers an efficient solution for 2D to 3D video conversion.
• This method effectively balances computational cost and visual fidelity.
• It presents a viable improvement for real-time 3D video synthesis applications.