Emission Ghost Imaging: reconstruction with data augmentation.

Area of Science:

• Optics and photonics
• Computational imaging
• X-ray imaging

Background:

• Ghost imaging reconstructs objects without spatial resolution using single-pixel detectors.
• Spatial modulation of incident beams with varying mask orientations is key to ghost imaging.
• Traditional methods face challenges with measurement matrix factorization when orientations are fewer than pixels.

Purpose of the Study:

• To improve ghost imaging reconstruction, especially when the number of mask orientations (N) is less than the number of pixels (P).
• To develop a robust method for 2D object reconstruction using ghost imaging techniques.
• To enhance the accuracy and reduce noise in ghost imaging reconstructions.

Main Methods:

• Developed a data augmentation method to enable QR factorization of the measurement matrix for N < P.
• Employed truncated singular value decomposition to determine the Moore-Penrose pseudoinverse for noise suppression.
• Applied the Adaptive Weights Smoothing method for further denoising of the reconstructed images.
• Utilized X-ray fluorescence data acquired at Brookhaven National Laboratory for experimental validation.

Main Results:

• The proposed method successfully performs QR factorization even when N < P.
• The combination of truncated SVD and Adaptive Weights Smoothing significantly reduces noise in reconstructions.
• Simulations show superior performance compared to Gram-Schmidt orthogonalization methods.
• Successful application to experimental X-ray fluorescence data demonstrates practical utility.

Conclusions:

• The novel data augmentation and denoising strategy enhances ghost imaging reconstruction accuracy and robustness.
• The method is particularly effective in scenarios with limited mask orientations.
• This work provides a significant advancement for ghost imaging applications, including X-ray fluorescence imaging.