Color computational ghost imaging based on a generative adversarial network.

Area of Science:

• Computational imaging
• Optical physics
• Machine learning in imaging

Background:

• Conventional ghost imaging often requires complex setups with multiple monochromatic paths.
• Noise and color distortion are significant challenges in ghost imaging reconstruction.
• Existing generative models may not fully capture subjective visual quality.

Purpose of the Study:

• To present a novel, simplified color computational ghost imaging scheme.
• To improve image reconstruction quality by reducing noise and color distortion.
• To achieve realistic and natural image textures with reduced data acquisition.

Main Methods:

• Utilized a single digital light projector for color speckle projection.
• Employed a single-pixel detector for light intensity measurement.
• Applied a generative adversarial network (GAN) for image noise and distortion removal.

Main Results:

• Successfully reconstructed color object images with simplified experimental setup.
• Achieved significant reduction in noise and color distortion compared to conventional methods.
• Demonstrated superior visual quality with realistic textures, even at a low sampling rate of 0.05.

Conclusions:

• The proposed color computational ghost imaging scheme offers a more efficient and effective approach.
• Generative adversarial networks are highly effective in enhancing image quality for ghost imaging.
• This method holds promise for applications requiring high-fidelity color image reconstruction with reduced complexity.