Computational ghost imaging based on a conditional generation countermeasure network under a low sampling rate.

Area of Science:

• Computational imaging
• Deep learning for optics
• Image reconstruction

Background:

• Traditional ghost imaging requires high sampling rates, limiting its efficiency.
• Deep learning approaches show promise but often demand extensive training data.
• Sub-Nyquist sampling presents challenges in reconstructing high-fidelity images.

Purpose of the Study:

• To develop an end-to-end deep neural network for computational ghost imaging.
• To achieve high-quality image restoration at sub-Nyquist sampling rates.
• To reduce the data requirements for ghost imaging reconstruction.

Main Methods:

• Proposed a conditional generative adversarial network for computational ghost imaging (CGANCGI).
• Input: 2D light signal from CMOS camera and object's gray image.
• Trained the CGANCGI network to recover object images directly from light signals.

Main Results:

• Achieved fast image restoration using only 1/10 of traditional deep learning samples.
• Demonstrated significant improvements in image quality: Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity (SSIM) were 4-6x and 5-7x higher, respectively, than the original image.
• Verified practical application prospects in ghost imaging under low sampling rates.

Conclusions:

• The CGANCGI method enables high-quality image recovery in ghost imaging with significantly reduced sampling rates and data needs.
• The approach offers a practical solution for efficient ghost imaging, particularly in scenarios with limited data acquisition.
• This deep learning-based technique advances computational ghost imaging by overcoming traditional sampling rate limitations.