Single-pixel imaging based on self-supervised conditional mask classifier-free guidance.

Area of Science:

• Optics and Photonics
• Computational Imaging
• Machine Learning

Background:

• Single-Pixel Imaging (SPI) aims for high-quality image reconstruction with minimal data acquisition.
• Current deep learning methods for SPI are limited by optimizing direct image reconstruction, restricting low measurement rate potential.
• Conditional probability and guidance models offer new avenues for improving reconstruction fidelity.

Purpose of the Study:

• To develop an advanced reconstruction method for Single-Pixel Imaging (SPI) that overcomes limitations of current deep learning approaches.
• To enhance image reconstruction quality under extremely low measurement rates.
• To introduce a novel self-supervised learning framework for improved SPI performance.

Main Methods:

• Proposed a self-supervised conditional masked classifier-free guidance (SCM-CFG) model for single-pixel reconstruction.
• Utilized conditional probability and classifier-free guidance (CFG) principles for enhanced reconstruction.
• Implemented a conditional mask design to improve overlay accuracy in image reconstruction.

Main Results:

• Achieved an average Peak Signal-to-Noise Ratio (PSNR) of 26.17 dB on the MNIST dataset at a 10% measurement rate.
• Demonstrated superior performance compared to existing photon imaging and computational ghost imaging methods.
• Showcased significant generalization capabilities and an average improvement of 7.3 dB in overlay processing accuracy.

Conclusions:

• SCM-CFG effectively reconstructs high-quality images from low-rate measurements in Single-Pixel Imaging.
• The proposed method surpasses current state-of-the-art techniques in both reconstruction accuracy and generalization.
• Physical experiments confirmed the practical effectiveness and advantages of the SCM-CFG approach.