Wrapped phase denoising: a WISE-transformer and comprehensive comparisons

Area of Science:

• Optical Metrology
• Deep Learning
• Signal Processing

Background:

• Wrapped phase denoising is crucial in optical metrology.
• Existing methods include traditional and deep learning approaches.
• There is a need for advanced denoising techniques with improved performance and understanding.

Purpose of the Study:

• To introduce WISE-Transformer, a novel U-shaped transformer model for wrapped phase denoising.
• To compare WISE-Transformer against traditional methods (WFT) and theoretical limits (CRB).
• To analyze the effectiveness of specific network inputs and data generation strategies.

Main Methods:

• Development of WISE-Transformer, a U-shaped transformer incorporating windowed and inter-window self-attention.
• Extensive experiments evaluating performance across multiple metrics.
• Comparative analysis against Windowed Fourier Transform (WFT) and Cramer-Rao Lower Bound (CRB).
• Verification of sine/cosine phase inputs and dynamic data generation.

Main Results:

• WISE-Transformer achieves state-of-the-art performance compared to other deep learning models.
• The model outperforms WFT and breaks the CRB in simulated environments.
• Sine/cosine phase inputs and dynamic data generation enhance training diversity and performance.

Conclusions:

• WISE-Transformer demonstrates superior performance in simulated wrapped phase denoising.
• The model shows competitive results on real experimental data but faces generalization challenges.
• Future research should focus on improving the generalization capabilities of deep learning denoising methods.