Wrapped phase denoising: a WISE-transformer and comprehensive comparisons.

Area of Science:

• Optical Metrology
• Computer Vision
• Signal Processing

Background:

• Wrapped phase denoising is crucial in optical metrology.
• Existing methods include traditional algorithms and deep learning approaches.
• There is a need for improved denoising techniques balancing local and global information.

Purpose of the Study:

• To introduce WISE-Transformer, a novel U-shaped transformer for wrapped phase denoising.
• To compare WISE-Transformer against traditional and theoretical benchmarks.
• To investigate input representations and data augmentation strategies.

Main Methods:

• Developed WISE-Transformer, a U-shaped transformer with windowed and inter-window self-attention.
• Utilized sine and cosine phases as network input.
• Employed a dynamic data generation strategy for training.
• Conducted extensive experiments comparing against Windowed Fourier Transform (WFT) and Cramer-Rao Lower Bound (CRB).

Main Results:

• WISE-Transformer achieved state-of-the-art performance on multiple metrics.
• The model outperformed WFT and broke the CRB in simulated environments.
• Sine and cosine phase inputs and dynamic data generation enhanced training.
• Performance on real experimental data was competitive but less effective than WFT.

Conclusions:

• WISE-Transformer represents a significant advancement in deep learning for wrapped phase denoising.
• The study highlights the potential of transformer architectures in this field.
• Generalization to real-world data remains a key challenge for future research.