High-Resolution Single-Pixel Imaging of Spatially Sparse Objects: Real-Time Imaging in the Near-Infrared and Visible...

Area of Science:

• Optics and Photonics
• Computational Imaging
• Machine Learning Applications

Background:

• Single-pixel imaging (SPI) offers a cost-effective alternative to traditional cameras, particularly for non-visible wavelengths.
• Existing SPI methods often face limitations in resolution, speed, and reconstruction complexity.
• Digital Micromirror Devices (DMDs) provide a versatile platform for optical modulation in SPI systems.

Purpose of the Study:

• To develop and validate a high-resolution single-pixel imaging framework for dynamic scenes.
• To optimize the sampling scheme and reconstruction algorithm for efficient data acquisition and processing.
• To evaluate the performance of iterative versus neural network-based reconstruction methods.

Main Methods:

• Implemented a novel SPI framework utilizing a digital micromirror device (DMD) at its native 1024x768 resolution.
• Developed a two-stage reconstruction algorithm involving generalized inverse matrix multiplication and comparative analysis of iterative and neural network approaches.
• Operated the system in visible and near-infrared wavelengths with a compression ratio of 0.41% and a measurement rate of 6.8 Hz.

Main Results:

• Achieved high-resolution SPI (1024x768) in visible and near-infrared spectra.
• Demonstrated real-time reconstruction comparable to the image acquisition rate using a desktop GPU.
• Neural network reconstruction excelled for similar training data, while iterative methods offered broader applicability.

Conclusions:

• The proposed SPI method enables high-resolution, dynamic imaging with standard hardware and real-time processing capabilities.
• The optimized framework supports diverse applications requiring rapid acquisition and analysis of sparse scenes.
• The study highlights the trade-offs between iterative and neural network reconstruction for SPI.