Single-pixel image reconstruction using coherent nonlinear optics.

Area of Science:

• Optoelectronics
• Image Reconstruction
• Deep Learning

Background:

• Reconstructing 2D images from noisy data presents significant challenges.
• Existing methods often struggle with noise contamination and limited data.

Purpose of the Study:

• To propose and demonstrate a novel hybrid optoelectronic system for reliable 2D image reconstruction.
• To optimize image reconstruction by maximizing multi-scale structural similarity index measure (MS-SSIM) and minimizing mean absolute error (MAE).

Main Methods:

• Utilizing mode-selective frequency upconversion and single-pixel detection.
• Employing a deep neural network for image reconstruction from a noise-contaminated database.
• Training the system to enhance image quality metrics like peak signal-to-noise ratio (PSNR) and structural similarity index measure (SSIM).

Main Results:

• Achieved reliable reconstruction of 2D images from handwritten digits with high fidelity.
• Obtained a peak signal-to-noise ratio (PSNR) of approximately 20 dB.
• Reached a structural similarity index measure (SSIM) of around 0.85, demonstrating effective image reconstruction.

Conclusions:

• The proposed hybrid optoelectronic system offers a robust solution for 2D image reconstruction in the presence of noise.
• The system's performance metrics indicate high-quality and faithful image reproduction.
• The technique's versatility makes it applicable to diverse fields such as Lidar, compressive imaging, and volumetric reconstruction.