Unsupervised speckle denoising in digital holographic interferometry based on 4-f optical simulation integrated...

Area of Science:

• Optics and Photonics
• Artificial Intelligence
• Image Processing

Background:

• Speckle noise in digital holographic interferometry (DHI) is an inherent challenge that compromises measurement accuracy.
• Existing denoising methods often struggle with the complexity and variability of speckle noise, limiting their effectiveness.

Purpose of the Study:

• To develop and evaluate a self-supervised deep learning approach for mitigating speckle noise in DHI.
• To enhance the accuracy and reliability of DHI measurements by effectively removing speckle noise.

Main Methods:

• A cycle-consistent generative adversarial network (GAN) was employed for speckle denoising.
• The method incorporates a 4-f optical speckle noise simulation module and a parameter generator.
• Training utilized an unpaired dataset, overcoming the limitations of acquiring noise-free and paired experimental data.

Main Results:

• The proposed deep learning method demonstrated superior speckle denoising performance on both simulated and experimental DHI data.
• Achieved a 6.9% performance improvement over conventional methods and a 2.6% improvement over unsupervised deep learning in peak signal-to-noise ratio (PSNR).

Conclusions:

• The self-supervised deep learning method offers a robust and effective solution for speckle noise reduction in DHI.
• The approach shows significant potential for improving DHI applications, especially in processing large datasets.