Passive Photon Counting

One can take pictures without a flash under low-light conditions by time-lapse photography, that is, by using a long exposure time to compensate for the scarcity of light. The price paid for this is blurring of any object that is moving. This trade-off between the ability to "see in the dark" and the ability to freeze motion applies whether one is using photographic film or electronic imaging devices. If the electronic imaging device adds noise to the image, more light must be collected to overcome this noise, and this translates into longer exposure time and more blur.

In a conventional electronic imager, such as a charge-coupled device (CCD) or a CMOS pixel sensor, the light is converted in each pixel to an electrical charge. When the image is read out, an amplifier circuit senses this photocharge and converts it to a usable output voltage. Like any analog circuit, this amplifier adds noise. This so-called readout noise becomes more severe when pixels are sensed frequently in an attempt to freeze motion. Figure 1 below illustrates the resulting trade-off.

Short vs long exposure issuesFigure 1. Left: Short exposure resulting in readout noise; right: long exposure resulting in motion blur.


A Geiger-mode avalanche photodiode (APD), on the other hand, can be used to build an all-digital pixel in which the arrival of each photon triggers a discrete electrical pulse. The photons are counted digitally within the pixel circuit, and the readout process is therefore noise-free. At low light levels, there is still noise in the image because photons arrive at random times so that the number of photon detection events during an exposure time has statistical variation. This noise is known as shot noise. One advantage of a pixel that can digitally count photons is that if shot noise is the only noise source, the image quality will be the best allowed by the laws of physics.

Another advantage of an array of photon counting pixels is that, because of its noiseless readout, there is no penalty associated with reading the imager out frequently. If one reads out a thousand 1-ms exposures of a static scene and digitally adds them, one gets the same image quality as a single 1-s exposure. This would not be the case with a conventional imager that adds noise each time it is read out. Now suppose the scene is moving, for example, because the images are being acquired from a moving platform such as an unmanned air vehicle. The digital summation of the 1-ms exposures can be done with preprocessing to compensate for the motion. In other words, the individual frames have enough information that they can be realigned before being added.

Digital image processingFigure 2. Digitally adding multiple short exposures to produce a higher quality image.


The figure above illustrates this idea. Using digital processing to compensate for the motion, one can get the best of both worlds: the light collection of a long exposure time with negligible motion blur.

With digital photon counting one can envision building "smart" imagers that can perform a variety of image processing functions that extract important information from the scene. Some of these functions could be done during the exposure time to reduce the amount of data that needs to be read out, thereby also reducing electronic bandwidth and power dissipation. Such an imager might be used to detect edges in a scene or to selectively pick out features, such as moving vehicles, that are changing in time.

Photon counting cameraFigure 3. Left: Photon-counting camera; right: one of the first images collected by the camera.


Lincoln Laboratory recently demonstrated a photon-counting camera based on a 256 × 256-pixel Geiger-mode APD array integrated to a CMOS readout chip with a digital counter in each pixel. On the left in the figure above is the camera system and on the right is one of the first images it collected.



top of page