Source-linked AI summary
Imaging with a small number of photons
Peter A. Morris, Reuben S. Aspden, Jessica Bell, Robert W. Boyd, Miles J. Padgett
TL;DR
Low-light imaging needs methods that recover useful images from very few photons while limiting illumination. The paper combines correlated-photon heralding with a time-gated ICCD and sparse-data reconstruction, producing images with fewer than one detected photon per pixel.
Problem
Low-light applications need imaging at reduced photon flux, while scanning-detector ghost imaging has detection efficiency limited to 1/N for an N-pixel image.
Method
The system uses SPDC-correlated photon pairs, a heralding SPAD, a time-gated multipixel ICCD, and reconstruction exploiting Poissonian data and spatial-frequency sparsity.
Results
Images of a USAF test target are reconstructed from 7000 detected photons, while a wasp-wing image uses 0.45 photons per pixel.
Takeaways & Limitations
Ghost imaging forms the image from photons that did not interact with the object, while heralding supports high-contrast imaging and virtually eliminates background counts.
Abstract
from arXiv · showhide
Low-light-level imaging techniques have application in many diverse fields, ranging from biological sciences to security. We demonstrate a single-photon imaging system based on a time-gated inten- sified CCD (ICCD) camera in which the image of an object can be inferred from very few detected photons. We show that a ghost-imaging configuration, where the image is obtained from photons that have never interacted with the object, is a useful approach for obtaining images with high signal-to-noise ratios. The use of heralded single-photons ensures that the background counts can be virtually eliminated from the recorded images. By applying techniques of compressed sensing and associated image reconstruction, we obtain high-quality images of the object from raw data comprised of fewer than one detected photon per image pixel.
INTRODUCTION
The paper develops camera-enabled low-light imaging that overcomes the efficiency limits of scanning-detector ghost imaging. It combines correlated photons, heralding, and sparse-data reconstruction to form images with fewer than one detected photon per pixel.
- INTRODUCTION: Scanning single-pixel ghost-imaging detectors limit detection efficiency to 1/N for an N-pixel image.Detector arrays can instead enable imaging with N-times fewer illumination photons.
- INTRODUCTION: Lower illumination flux is potentially valuable in biological imaging, where bleaching or sample damage can occur, and in covert security imaging.
- INTRODUCTION: The system characterises camera-enabled, time-gated imaging with the object placed either in the heralding arm or the camera arm.
- INTRODUCTION: Sparse-data reconstruction uses Poissonian statistics and an appropriate sparse domain to improve images from fewer than one photon per image pixel.The authors report images of a biological sample using fewer detected photons than image pixels.
EXPERIMENTAL METHODS
The imaging platform uses SPDC photon pairs, a heralding SPAD, and a time-gated ICCD camera. External triggering and optical-delay compensation align camera detections with correlated photon pairs across several imaging configurations.
- EXPERIMENTAL METHODS: A 355 nm laser pumps a type-I BBO crystal to generate near-collinear, frequency-degenerate photon pairs selected around 710 nm.The photons are separated by a pellicle beam splitter into camera and heralding arms.
- EXPERIMENTAL METHODS: A single-photon avalanche detector heralds the partner photon and triggers the multipixel ICCD camera.
- EXPERIMENTAL METHODS: The ICCD intensifier gate lasts several nanoseconds, while CCD exposure and readout occur over typically several seconds.
- EXPERIMENTAL METHODS: Each CCD frame accumulates all single-photon events recorded during its exposure.
- EXPERIMENTAL METHODS: An added optical path compensates electronic triggering delay so camera and herald detections correspond to the same photon pair.
- EXPERIMENTAL METHODS: The system compares ghost, heralded, and direct imaging by changing object placement and whether the camera receives heralded or internal triggers.In ghost imaging, the object is in the heralding arm; in heralded imaging, it is in the camera arm.
IMAGE ACQUISITION
Images are acquired in photon-sparse frames and compared across ghost, heralded, and direct configurations. Heralded triggering produces clear, high-contrast images, whereas direct periodic triggering yields only a faint image.
- IMAGE ACQUISITION: 900 frames of 2 s exposure are summed, with the intensifier firing on each heralding-detector or internal trigger.
- IMAGE ACQUISITION: The 600 × 600-pixel region of interest covers (7.8×7.8) mm^2, and acquisition is chosen to remain much less than one photon event per pixel per frame.
- IMAGE ACQUISITION: The measured camera-readout dark-count probability is 5 × 10^-4 per frame per pixel.Photon counting uses a binary threshold calibrated from 100 triggered frames acquired with the shutter closed.
- IMAGE ACQUISITION: GI and HI both produce clear target images with contrast of order 40:1, while DI produces only a very faint image.
- IMAGE ACQUISITION: In DI, random photon arrivals only occasionally coincide with the regularly firing intensifier window, reducing detection efficiency and losing coincidence behavior.
OPTIMISATION OF RECONSTRUCTED IMAGE
The reconstruction selects statistically plausible images by balancing Poisson log likelihood against sparsity in the spatial-frequency domain. Adjusting λ produces a trade-off between preserving fine structure and smoothing the image, enabling reconstruction below one photon per pixel.
- Reconstruction principle: Compressed sensing exploits spatial-frequency sparsity while maintaining consistency with Poissonian photon statistics.
- Reconstruction principle: The optimisation maximises a merit function combining reconstructed-image log likelihood with a spatial-frequency participation penalty.
- Regularisation trade-off: λ sets the balance between satisfying recorded data and enforcing sparsity; low values preserve sparse data, whereas high values produce overly smooth images.
- Photon-limited imaging: < 7000 photons produced a reconstructed USAF test-target image, corresponding to less than 0.2 photons per image pixel.
- Photon-limited imaging: 40419 detected photons formed a wasp-wing image over 90000 pixels, corresponding to 0.45 photons per pixel.
CONCLUSIONS
The study develops time-gated low-light imaging that combines photon counting, ghost imaging, and sparsity-based enhancement. It reconstructs images from fewer than one photon per pixel, including a wasp-wing image at 0.45 photons per pixel.
- Conclusions: A camera-enabled, time-gated system combines photon counting with spatial-frequency sparsity and Poissonian image enhancement.
- Conclusions: 7000 detected photons reconstructed a USAF test target, while the imaging techniques enabled photon numbers below one photon per pixel.
- Conclusions: A time-gated ghost-imaging configuration acquired a wasp-wing image with an average ratio of 0.45 photons per pixel.