Abstract
Imaging in near-total darkness presents significant challenges for conventional sensors due to read noise and sensitivity limits. This paper presents an Imaging System based on Single Photon Counting and Compressive Sensing (ISSPCCS) designed to reconstruct sparse images in absolute darkness. By integrating a single-photon avalanche detector (SPAD) with a digital micromirror device (DMD) and utilizing convex optimization algorithms, the system achieves high-fidelity image reconstruction from under-sampled data. Experimental results demonstrate successful imaging with as few as 34 photons per pixel per second, offering a robust and intuitive alternative to traditional analog single-pixel cameras.
Introduction
Capturing high-quality images in "ultra-dark" environments is a persistent hurdle in optical engineering. Standard CMOS or CCD sensors are limited by thermal and read noise, which often exceeds the weak signal from a few incident photons. While single-photon detector arrays (SPAD arrays) offer a solution, they are expensive and difficult to manufacture at high resolutions.
The ISSPCCS addresses these issues by combining the high sensitivity of a single-pixel Geiger-mode avalanche photodiode (APD) with the efficiency of compressive sensing (CS). Unlike traditional single-pixel cameras that use photomultiplier tubes (PMTs) for analog intensity measurements, the ISSPCCS counts discrete photon events, providing a more robust signal in extreme low-light conditions.
System Architecture and Methodology
The hardware setup of the ISSPCCS consists of three primary modules: a spatial light modulator, a high-sensitivity detector, and a reconstruction framework.
- Digital Micromirror Device (DMD): This serves as the spatial light modulator. It consists of an array of micro-mirrors that can be independently tilted to "on" or "off" positions, applying pseudorandom binary patterns to the light reflected from the object.
- Geiger-mode Silicon APD: This single-pixel detector triggers a digital pulse for every detected photon. This discrete counting mechanism is inherently more intuitive and less noisy than analog measurements in ultra-weak light.
- Compressive Sensing (CS) Framework: CS exploits the inherent sparsity of most natural images. Instead of following the Nyquist-Shannon sampling theorem, the system takes $M$ measurements (where $M$ is significantly less than the total number of pixels $N$) and reconstructs the image using the SpaRSA (Sparse Reconstruction by Separable Approximation) algorithm.
Experimental Results and Analysis
The system was evaluated using physical targets, specifically films printed with the abbreviations "NSSC" and "BIT."
Sensitivity and SNR
The ISSPCCS demonstrated remarkable sensitivity, producing clear images of the "NSSC" target (64 × 256 pixels) with an average light intensity of only 34 photons per pixel per second. Despite the extreme darkness, the system achieved a signal-to-noise ratio (SNR) of 12.2 dB.
Efficiency and Robustness
The power of compressive sensing is evident in the system's ability to reconstruct images from under-sampled data. For a 4,096-pixel area, successful reconstructions were achieved with only 1,800 measurements. Even when the number of measurements was further reduced to 600, the target letters remained recognizable, showcasing the robustness of the SpaRSA algorithm.
High-Resolution Scanning
To overcome computational limits for larger images, the authors implemented a "DMD work area scanning" method. By software-controlling the active region of the DMD to scan row-by-row, they successfully reconstructed a 1024 × 79 pixel image of the "BIT" target.
Discussion and Conclusion
The ISSPCCS provides a compact (under 0.2 m²) and highly sensitive solution for low-light imaging. By shifting from analog intensity measurements to discrete photon counting, the system gains significant robustness against noise.
However, limitations remain regarding temporal resolution. The current DMD operating speed (34 Hz) results in long acquisition times (e.g., ~53 seconds for 1,800 measurements), making the system unsuitable for dynamic scenes. Future work will focus on increasing the modulation speed and optimizing reconstruction algorithms for real-time applications. Potential use cases include biological fluorescence imaging and non-destructive material inspection in light-sensitive environments.
Reference: Yu, W. K., et al. (2012). Single photon counting imaging system via compressive sensing. Optics Express.
Summary updated