Learning Non-Local Spatial Correlations To Restore Sparse 3D Single-Photon Data.

Area of Science:

• Photon-starved imaging
• 3D Lidar data processing
• Computational imaging

Background:

• Single-photon 3D Lidar imaging is crucial for various applications but suffers from noise and low data density.
• Existing restoration algorithms struggle with photon-starved regimes and sparse spatial sampling.
• Accurate depth and reflectivity estimation is vital for reliable 3D scene reconstruction.

Purpose of the Study:

• To develop a new algorithm for learning spatial correlations and restoring 3D Lidar images under photon-limited conditions.
• To improve the quality of depth and reflectivity images acquired with sparse spatial sampling.
• To address the challenges posed by the photon-starved regime in 3D Lidar data.

Main Methods:

• A novel algorithm combining multi-scale information extraction, non-local spatial correlation graph construction, and image restoration.
• Implementation of a non-uniform sampling strategy to optimize computational cost.
• Efficiently solving the image restoration problem using the alternating direction method of multipliers (ADMM).

Main Results:

• The proposed algorithm significantly enhances the quality of estimated depth and reflectivity images.
• Demonstrated benefits in both simulated and real 3D Lidar data.
• Effective restoration achieved even in photon-starved conditions or with reduced spatial points.

Conclusions:

• The developed algorithm offers a robust solution for restoring 3D Lidar images with limited data.
• It provides improved accuracy for depth and reflectivity estimation, crucial for 3D reconstruction.
• The method is particularly valuable for applications requiring high-quality 3D Lidar data under challenging acquisition constraints.