Example of a D200 Photosite Analysis
Prepared 2006-05-30 (149/33701) by Bill Claff

For some time I have been performing a statistical analysis of the linear data in NEF files provided by Nikonians to help them document bad photosites.
I have no complaint about my D200 sensor but I thought it would be useful to present an analysis as an example.
BTW, this example is far more extensive than is typical since I generally only receive one image to analyse from Nikonians.
(E-mail me if you want such an analysis.)

I'm sorry for the long post but believe it or not this is the compressed version

Preface (for geeks):
The analysis makes the assumption that noise at the test exposures is dominated by Photon Noise (also known as shot noise).
Therefore there is an expectation that the data for a given channel will behave like a normal (actually Poisson) distribution.
This means that all the data should fall within a certain number of standard deviations from the mean.
Given 2,509,056 photosites per channel at least 5-6 standard deviations are expected due to photon noise alone for the D200.

I took a series of images of a white computer screen at every ISO from 100 to 3200.
I used a 50mm f/1.8D lens in Aperture mode at f/8 with Exposure Compensation of +2EV.
The body was in Manual Focus, the lens was focused at infinity, and the front of the lens was against the computer screen.
The technique produces well exposed, very flat, "white" images.

Lens cap tests (for geeks):
Are not good.
They are dominated by Dark Noise and are not representative of "ordinary" photography.
Frequently the "bad" photosites located are within the bounds of ordinary noise.

The linear data was analysed and the initial results are shown as Gb, B, R, and Gr on the attached chart.

First note the magenta line for R; clearly the Red channel is not performing as well as the others even at ISO 100.
I located two Red photosites that are slightly warm.
(At a Signal to Noise Ratio of 5.5, I certainly don't consider then hot or stuck.)
When these two Red photosites are excluded from the data the red line labeled R(-2) results.

Now note the cyan line for B; clearly at about ISO 250 at least one Blue photosite starts to misbehave.
I located two Blue photosites affecting ISOs 250 and up as well as one that affects only ISO 3200.
I have not seen these in any of my images (yet) but given my experience I think they might show up at ISO 1250 when the become more than 10 standard deviations too high.
When these three Blue photosites are excluded from the data the blue line labeled B(-3) results.

Notes (for geeks):
The slight upward trend is due to the use of Aperture mode.
As the exposures get shorter Read Noise makes more of a contribution to the overall noise.
(Actually Photon Noise makes less of a contribution.)
The red blip at ISO 1600 appears to be a truly random single photosite event.

Clearly the bad photosites that really matter are those that affect our photography.
Never-the-less I take comfort in having an objective way of locating bad photosites and measuring their "badness".