RS-N2N: a single-image phase denoising network for digital holographic microscopy

Area of Science:

• Optics and Photonics
• Biomedical Imaging
• Artificial Intelligence in Imaging

Background:

• Phase noise in digital holographic microscopy (DHM) severely impacts measurement accuracy.
• Existing deep learning denoising methods require extensive paired datasets and struggle with real-world noise generalization.

Purpose of the Study:

• To develop a novel single-frame denoising network for DHM phase maps.
• To improve the robustness and effectiveness of noise removal in DHM.

Main Methods:

• Proposed RS-N2N, a single-frame denoising network incorporating Fourier-based preprocessing and SE-block attention.
• Implemented a self-constrained learning strategy with data augmentation for training on single noisy images.

Main Results:

• RS-N2N achieved 1.3 dB higher PSNR than previous methods for Gaussian noise.
• Demonstrated up to 3 dB improvement for Perlin noise and an average of 1.5 dB gain for real phase noise.
• Outperformed existing methods in denoising phase noise for DHM.

Conclusions:

• RS-N2N effectively denoises phase noise in DHM using single-frame images.
• The proposed method offers a practical solution for enhancing DHM measurement accuracy.
• RS-N2N shows superior performance compared to conventional denoising techniques.