Depth Map Restoration From Undersampled Data.

Area of Science:

• Computer Vision
• Image Processing
• 3D Reconstruction

Background:

• Low-cost sensors yield low-resolution depth maps.
• Structure from motion and sparse scanning produce sparse point clouds.
• Depth map restoration requires uniform or non-uniform upsampling.

Purpose of the Study:

• To propose a unified framework for depth map restoration using sparse representation.
• To address both uniform upsampling of low-resolution depth maps and non-uniform upsampling of sparse depth maps.
• To develop a method that does not rely on high-resolution intensity images.

Main Methods:

• Utilizing sparse representation and exemplar-based sub-dictionaries for depth map restoration.
• Implementing an edge-preserving constraint and pyramidal strategy for uniform upsampling.
• Employing exemplar similarity for non-uniform upsampling and reconstructing missing depth information.
• Suggesting sequential cascading of uniform and non-uniform upsampling for very sparse data.

Main Results:

• Demonstrated efficacy in restoring high-resolution and dense depth maps.
• Successfully handled both uniform and non-uniform upsampling requirements.
• Showcased the ability to reconstruct depth maps without high-resolution intensity images.
• Validated through diverse qualitative and quantitative results.

Conclusions:

• The proposed sparse representation framework effectively restores high-resolution depth maps.
• The unified approach handles diverse depth map restoration challenges, including uniform and non-uniform upsampling.
• The method offers a viable alternative for depth map enhancement without auxiliary intensity data.