How close are digital sensors to “night vision” image intensifiers?

Question

Traditionally reserved for military applications, image intensifier tubes (a.k.a. night vision scopes) are persistently pricey analog devices for amplifying light.

For non-military applications they are largely being overtaken by "digital night-vision," which is essentially a conventional sub-$100 CMOS or CCD sensor without an infrared filter, typically supplemented with active near-infrared illuminators.

Given that a state-of-the-art image intensifier (which may be called 4G, or Gen3+) starts at $3,000, I am wondering: How close are existing digital camera sensors to matching the performance of analog intensifiers, and is there a theoretical limit that would prevent a conventionally-sized CMOS or CCD sensor from matching the low-light performance of an analog intensifier?

I'm not sure how to compare the performance specifications provided for analog intensifiers to a "black-and-white" digital sensor. For example, state-of-the-art intensifier specifications include:

• 64lp/mm (line pairs per mm) resolution
• Signal/Noise about 25.

Note that for "night vision" a monochromatic digital sensor can eschew the anti-aliasing and color filter arrays, presumably increasing sensitivity and reducing noise over the conventional ratings we see with those filters in place. DxO offers a low-light sensor measurement that lists "the highest ISO setting for a camera that allows it to achieve an SNR of 30dB while keeping a good dynamic range of 9 EVs and a color depth of 18bits." (SNR of 30dB equates to a simple multiple of 32. And the color requirements are, I assume, far more demanding than those required for black-and-white resolution.) For current generation full-frame sensors those numbers are up around 3000ISO. For APS-C sensors it looks like they're around 1000ISO.

Perhaps someone with a greater facility for these metrics can find a relationship between the analog and digital ones?

Answer 1

This is a nontrivial question (and I went thru this comparison years ago while designing high-speed sensors for adaptive optics systems). All the intensifiers I'm familiar with, such as microchannelplate and high-voltage image tubes have rather low quantum efficiency compared with solid-state sensors. So while you get a ton of output photons for each detection event, you essentially get amplified Poisson shot noise in the image.

The solid-state sensor, then has to deal with the internal analog noise (in readout, preamps, etc) prior to the A/D converters. A good-quality sensor will have noise under one electron per readout, tho' those are rather more expensive than you'll get in a consumer camera. Note that this can be for an integration time of seconds.
There's a ton of "specmanship" in the values advertised for cameras, and sadly it's hard if not impossible to dig up the actual Q.E. and analog noise floor values.

By comparison, the cameras w/ active IR illuminators (such as "Trail Cams") do produce pretty nice images for under $150. I have one set up, and other than massive reflection from animals' retinas, the shots are clear and noise-free. But I don't count any illuminated image as "low-light camera" :-) .