Super-Resolution Phase Retrieval Network for Single-Pattern Structured Light 3D Imaging.

Area of Science:

• Computer Vision
• 3D Imaging
• Optical Metrology

Background:

• Structured light 3D imaging is crucial for accurate 3D data acquisition using phase retrieval.
• Single-pattern methods offer speed advantages over multi-pattern techniques but suffer from lower accuracy and resolution limitations.
• Improving projector resolution in existing systems is cost-prohibitive.

Purpose of the Study:

• To develop a novel super-resolution phase retrieval network (SRPRNet) to address the accuracy and resolution challenges in single-pattern structured light 3D imaging.
• To enable high-resolution 3D reconstruction from single-pattern inputs, reducing hardware costs.
• To provide a versatile solution applicable to both standard and super-resolution phase retrieval tasks.

Main Methods:

• Proposed SRPRNet incorporates a phase-shifting module for multi-scale feature extraction with varying phase shifts.
• A refinement and super-resolution module generates enhanced, high-resolution phase components.
• Introduced sine and cosine shifting losses as regularization terms for the loss function.

Main Results:

• SRPRNet demonstrates state-of-the-art performance for $1\times $, $2\times $, and $4\times $ super-resolution phase retrieval.
• Experimental validation across three datasets confirms the network's effectiveness.
• The method successfully achieves high-resolution absolute phase after demodulation and unwrapping.

Conclusions:

• SRPRNet is the first network capable of super-resolution phase retrieval using a single structured light pattern.
• The proposed network significantly improves accuracy and resolution in single-pattern 3D imaging.
• SRPRNet offers a cost-effective solution for high-fidelity 3D reconstruction.