Deep learning based coherence holography reconstruction of 3D objects.

Area of Science:

• Optical physics
• Computational imaging
• Machine learning

Background:

• Coherence holography is a powerful technique for 3D object reconstruction.
• Traditional analytical methods like Fourier fringe and sin-fit algorithms face limitations in accuracy, resolution, and reconstruction time, especially with noisy data.

Purpose of the Study:

• To develop a novel reconstruction method for coherence holography using deep neural networks.
• To enhance the accuracy, resolution, and efficiency of 3D object reconstruction from interferograms.

Main Methods:

• Development of deep neural network models, specifically conditional Generative Adversarial Networks (cGAN) and U-NET.
• Application of these models, termed deep coherence holography (DCH), to predict non-diffracted fields or sub-objects from interferograms.

Main Results:

• DCH achieves superior accuracy, resolution, and reconstruction speed compared to traditional methods.
• DCH requires only one image per sub-object, reducing total reconstruction time by N×.
• DCH demonstrates significantly lower mean square error (MSE) and higher peak signal-to-noise ratio (PSNR) for both amplitude and phase, especially with noisy interferograms.
• Reconstruction resolution is comparable to sin-fit and twice as good as Fourier fringe analysis.

Conclusions:

• Deep coherence holography (DCH) offers a significant advancement in 3D object reconstruction for coherence holography.
• The proposed deep learning approach provides a more robust and efficient alternative to conventional analytical imaging techniques.