2017-12-20 by Bill Claff
purpose of this document is to introduce people to the general guidelines for
collaborating in gathering sensor characterization data.
I gather raw image files of several particular styles and perform statistical analysis on those files to present findings.
Full size raw images are required for the analysis.
Set all noise
reduction settings to off (or a low as possible if off is not an option). In
particular do not use High ISO Noise Reduction.
DR boost if available should be off and any lens corrections disabled if possible.
The desire is to get as "pure" a raw file as possible.
For Sony uncompressed is preferred to eliminate any artifacts from the lossy compression.
If there is both a mechanical and electronic shutter use the mechanical.
If your camera does more than one bit depth, such as 14‑bit and 12‑bit then use the higher bit depth.
You will be using
Manual mode and Manual focus. Make sure Exposure Compensation is 0.
I always test every ISO including any intermediate or extended ones.
On some cameras you will need to change something to enable extended ISO settings.
If your camera does 1/3 EV settings then change the menu if necessary to allow that as well.
I have constructed these tests as a compromise between what one might do in a laboratory, with the freedom to take a large number of images and control a number of variables, and what is practical to ask ordinary photographers to provide.
I typically collect three sets of data.
of the images are simply black frames.
Generally these are taken with the lens (or body) cap on and at the highest shutter speed. (eg. 1/8000s) and smallest aperture (eg. f/22).
Body cap is best because it defeats any lens processing that might occur.
It's also not a bad idea to cover the viewfinder (if there is one) to guard against light leakage.
Take a single black frame at every ISO setting including extended and intermediate ISOs.
Choose any lens you like. Probably it will be easier with a short and light lens.
will be taken at the lowest analog ISO.
For most cameras the lowest numbered ISO setting is the lowest analog ISO; but on some its the next ISO setting up.
When in doubt wait for my feedback based on the Read Noise images.
Simply take 16 black frames as you did for Read Noise (set#1).
exposed frames are taken of your computer screen at the same ISO setting as the
You will set your lens to infinity focus and place the front of the lens (filter threads, not you actual lens!) against the computer screen.
Some people prefer to use a tripod, the important thing is that all the images are as equally exposed as is possible.
This ensures a diffuse image. Because some cameras modify raw data at low f-numbers I suggest f/4 as a good aperture.
Also, at f-numbers lower than 4 there is often light fall-off that interferes with the analysis.
A short primes lens is usually a good choice but wide angles lenses are a problem because they have such a deep Depth Of Field (DOF).
The following image is preferred because it exposes the red and blue channels more evenly than a simple white image:
can also use any other uniform target other than the computer screen but will
need a tripod for consistent images if the lens is not against the screen.
Take 16 of these images.
the sensor does dual conversion gain a second set of FPN Black and FPN
Illuminated frames is desirable.
I need to inspect the Read Noise files to determine which ISO setting to use for this additional data.
Contact me at BClaff@post.harvard.edu to gain access to my Dropbox to transfer your files or to arrange for an alternate method.