Computational ghost imaging enhanced by degradation models for under-sampling.

Area of Science:

• Optics and Photonics
• Computational Imaging
• Machine Learning in Imaging

Background:

• Computational ghost imaging (CGI) typically uses a single sampling ratio for 2D image acquisition.
• Existing CGI methods face limitations in reconstructing high-quality images from under-sampled measurements.
• Developing robust CGI techniques for efficient data acquisition is crucial for practical applications.

Purpose of the Study:

• To propose and validate a novel CGI method enhanced by degradation models for improved under-sampling.
• To leverage measurements from varying sampling ratios for robust image reconstruction.
• To advance the capabilities of CGI in scenarios with limited data acquisition.

Main Methods:

• Implementation of a CGI technique incorporating degradation models specifically designed for under-sampling scenarios.
• Utilizing results from both low-sampling-ratio and normal-sampling-ratio measurements to train a neural network.
• Employing self-supervised learning to fit the degradation model to the acquired data.
• Optimizing neural network parameters for enhanced image reconstruction.

Main Results:

• Experimental demonstration of improved 2D image reconstruction using the proposed CGI method with under-sampling degradation models.
• Successful training of a neural network capable of handling data from different sampling ratios.
• Validation of the method's effectiveness in reconstructing images from sparse measurements.

Conclusions:

• The proposed CGI method effectively enhances image reconstruction under under-sampling conditions.
• Degradation models integrated with neural networks offer a powerful approach for improving CGI performance.
• This advancement has the potential to significantly benefit various CGI applications requiring efficient data capture.