ClarkVision.com

To Clarkvision.com Galleries
Home Galleries Articles Reviews Best Gear Science New About Contact

Procedures for Evaluating Digital Camera
Sensor Noise, Dynamic Range, and Full Well Capacities;
How to Take the Data

All images, text and data on this site are copyrighted.
They may not be used except by written permission from Roger N. Clark.
All rights reserved.


Shine an incandescent light (or two) on a sheet of white paper, say a meter away (distance not critical). Be sure no fluorescent lights are shining on the paper. Have the incandescent light(s) warm up for at least a half hour.

Put the camera on a tripod, set up close to the paper so that only a small spot is imaged (1 to a few inches across). Use a moderate focal length lens, like 135 mm to 200 mm. A zoom lens is fine, and f/5.6 or f/8 is fine. Lens selection is not critical. Just avoid using a very wide angle or very fast apertures to minimize light fall-off toward the corners. Some light fall-off is OK and all lenses have it. Set the lens to infinity so the image is very out of focus. Turn off auto focus. Record images as raw. If the tripod or configuration slips during the imaging sequence you have to start over.

Set the camera on manual at its lowest ISO setting. If the camera has "extended range," turn it on and use the lowest ISO in the extended range. Run some test shots increasing exposure until the camera indicates the image is saturated, even in the corners. Delete those images. This is your longest and starting exposure. It should not be longer than 2 seconds. If it is move the light closer.

Now start a series of exposures, doing two shots at each exposure and keep the f/stop constant. After two frames at the same exposure, shorten the exposure by 1/3 (or 1/2 stop), and take two more frames. Keep doing that until the histogram shows the intensity level is well below saturation, then do pairs every stop until you reach the shortest exposure time. That provides the series needed to determine full well capacity and test if the camera is photon noise limited (it should be).

Next increase the ISO to the next ISO (e.g. 100 to 200). and take pairs of images just as you did in the above tests, starting with the longest exposure you used above. If some exposures are at the saturation level, that is OK, just continue to the shortest exposure the camera can do, like above.

The best data set would continue up in ISO doing the above test of pairs of exposures and various exposure times, increasing ISO one stop for each set. For example, sets at lowest ISO, ISO 100, 200, 400, 800, 1600. Higher ISOs do not give significant new data. Most information will be learned from the lowest ISO set, but higher ISOs might reveal other artifacts in the system. The minimum set is the low ISO sequence.

Next, find for the lowest ISO, an exposure that gives a bright response that is not saturated. Take a pair of exposures. Next increase the ISO 1 stop (e.g. 100 to 200), and reduce the exposure time one stop to compensate and take another pair of exposures. Be sure you are not saturating, and if so, erase those two exposures and reduce the exposure and make a new pair. Continue this sequence increasing the ISO by a stop each time up to the maximum ISO the camera will do.

Next is a series to test read noise. Again take pairs of images. In a relatively dark room with the lens cap on, take pairs of exposures at the fastest shutter speed (e.g. 1/4000 to 1/8000 second): a pair at each factor of 2 ISO (50, 100, 200, 400, 800, 1600, 3200, 6400, ...). Note the temperature of the room.

(Optional) Also for the bottom 3 ISOs: take pictures of any scene where something in the scene completely saturates the sensor. For example, bright clouds in a landscape scene. It can also be an indoor scene.

Some cameras clip the low end, while others have an offset to the digital data so when there is no light on the sensor, the data properly records the noise. If the camera clips the low end data, single pairs of fast exposures in a dark room to determine read noise is not adequate. Pairs of exposures just above zero level are needed and the pairs subtracted, noise statistics measured and the statistics modeled, projecting to zero light level. Thus for each ISO, pairs of exposures that give light levels from zero to a few percent of maximum signal are needed. It is not necessary to go all the way up to camera saturation. This sequence is a must for Nikon cameras because all Nikon cameras tested so far clip the low end. All canon cameras tested so far do not have this problem.

This series should take only 20 to 30 minutes, 30 to 40 minutes if the low end data are clipped at zero and you must take additional exposures at short exposure times (e.g. for Nikon).

If you still have time, some long exposure dark frames to test thermal noise. Again pairs of images, at ISO 1600. The test should be done in a dark room with the lens cap on. You'll need a remote timer, e.g. a TC-80N3 for many Canon DSLRs or a bulb setting where you can leave the camera exposing form minutes. Do a pair at 30 seconds, 1 minute, and 10 minutes (600 seconds). Note the temperature. If the timer does not work, or you don't have a bulb locking cable release, a pair of 30 second exposures at room temperature, iso 1600, could at least indicate something. Another set with the camera in a refrigerator and/or a freezer would help characterize the rate at which dark current changes with temperature, but only ISO 1600 and 600 second exposures are needed at cold temperatures.

It takes me about 5 days of work to analyze the data and write it up.


Back to the Procedures for Evaluating Digital Camera Sensor Noise, Dynamic Range, and Full Well Capacities main page.


http://www.clarkvision.com/articles/evaluation-1d2/howtotakedata.html

Home Galleries Articles Reviews Best Gear Science New About Contact

First published November 10, 2006.
Last updated December 5, 2014