It’s no secret that cameras are quickly getting better at capturing images in low light. Researchers at the University of Glasgow have pushed this trend to create an imager that can work with less than 1 photon per pixel. By combining two esoteric technologies — photon heralding and compressive imaging– the team has achieved a milestone that on the surface seems impossible. Leaving aside the huge amount of math and physics required under the covers, the process itself is actually fairly straightforward and very clever.
The first half of the weird science — heralded imaging, which is also called “ghost imaging” — is based on what are called heralded photons. Under certain circumstances pairs of quantum-entangled photons can be produced using a process called spontaneous parametric down-conversion and then split apart. Most of the time, when you detect one, the other one can also be detected. The detection of the first photon “heralds” the existence of the second.
The team’s imager uses a beam splitter to send one of each pair of photons it creates through the object being imaged and to a very sensitive single-pixel detector. It sends the other to a high-speed camera. The detector is only activated when a photon is sensed coming through the target object. When one does, the detector sends a signal to open the shutter of the camera — located at the end of the path of the other photon from the original pair — for about 15 nanoseconds. That’s long enough to record the position of the second — heralded — photon, but short enough to keep out almost all background noise. In essence, the single-pixel detector acts as a very high-speed shutter for the camera, so that it only takes pictures of photons that have passed through the target. To allow time for the shutter release signal to get from the detector to the camera, a delay line of about 70 nanoseconds is added to the photon’s path to the camera.
This use of heralded photons gets the imager’s light needs down to almost one photon per pixel — although there is still the unavoidable shot noise that comes with the Poisson distribution of photons. Compressive imaging allow the imager to deal with this noise, and to to push the boundaries even further – to less than one photon per pixel. By relying on the inherent redundancy of information in natural subjects, compressive imaging uses frequency domain information – in this case generated by performing a Discrete Cosine Transform on the image — to essentially reconstruct portions of the image that were not directly captured.
You can see the benefit of the compressive imaging technique in this pair of images provided by the researchers. The one on the left is an image of a wasp’s wing captured using about 10,000 photons. The one on the right is the reconstructed version after frequency information in the image is processed to create a “most likely” version of the original subject:
This amazing camera isn’t just for show. The team hopes it can lead to the development of cameras for use in science research that can be used to study and document subjects that are very light-sensitive, like certain biological specimens.