Fast single fringe-pattern processing with graphics processing unit.

Area of Science:

• Optical metrology
• Computational imaging
• High-performance computing

Background:

• Optical interferometric techniques offer noncontact, high-precision measurements crucial for various scientific fields.
• Single fringe-pattern processing (SFPP) is essential for analyzing fast phenomena but faces computational challenges due to large, high-resolution datasets.
• Existing SFPP methods struggle with processing time, hindering the analysis of dynamic events.

Purpose of the Study:

• To propose a general and complete graphics processing unit (GPU)-based framework for accelerating single fringe-pattern processing (SFPP).
• To systematically discuss and demonstrate SFPP acceleration through parallelization strategies.
• To establish a foundational framework for GPU-based SFPP applicable to diverse methods.

Main Methods:

• Development of a comprehensive GPU-based framework integrating denoising, phase demodulation, and phase unwrapping.
• Implementation of coherence-enhancing diffusion for noise removal.
• Utilization of spiral phase quadrature transform for phase demodulation and quality-guided phase unwrapping.

Main Results:

• The proposed GPU-based framework achieves significant acceleration for SFPP tasks.
• Experiments demonstrated a speedup of up to 29.8 times compared to traditional CPU-based sequential processing.
• The framework effectively reduces processing redundancy through analysis and parallelization.

Conclusions:

• The developed GPU-based SFPP framework offers a substantial advancement in processing speed and efficiency.
• This work represents the first complete GPU-based framework for SFPP, providing a valuable reference for parallelization strategies.
• The framework enables faster analysis of complex optical measurements, benefiting research in dynamic phenomena.