High-speed computational ghost imaging based on an auto-encoder network under low sampling rate.

Area of Science:

• Computational imaging
• Machine learning applications
• Optical sciences

Background:

• Traditional computational ghost imaging struggles with low sampling rates, limiting its practical applications.
• High-quality image reconstruction is crucial for various scientific and industrial fields.

Purpose of the Study:

• To propose a novel high-speed computational ghost imaging method.
• To enable accurate image reconstruction under low sampling conditions.
• To enhance image quality metrics like peak signal-to-noise ratio and structural similarity.

Main Methods:

• Development of an auto-encoder convolutional neural network architecture.
• Training the network for image reconstruction without requiring labeled datasets.
• Implementation of the auto-encoder for computational ghost imaging.

Main Results:

• Achieved significant improvements in peak signal-to-noise ratio (up to 18) and structural similarity (up to 0.7) under low sampling rates.
• Demonstrated high-quality image reconstruction with substantially fewer training samples (1/10) compared to traditional deep learning methods.
• Showcased the network's generalization capability for reconstructing gray-scale images.

Conclusions:

• The proposed auto-encoder based computational ghost imaging method offers a fast and effective solution for low sampling rate scenarios.
• This approach significantly enhances image reconstruction quality and reduces data requirements.
• The method shows promise for broader applications in computational imaging and machine learning.