D300 and D200
Do Noise Reduction on Longer Exposures

Prepared 2008-03-13 (149/118594) by Bill Claff

In a previous article I demonstrated that noise drops on
longer exposures even if all noise reduction settings are turned off.

For the D200 this effect starts at 1s and for the D300 it starts at 1/4s (and
the D3 too, but that's a different forum).

The analysis was performed on raw data, so obviously even NEF files are
affected.

There is some question of what is really going on.

Emil Martinec, a knowledgeable and frequent poster elsewhere, used a 2D Fast
Fourier Transform (FFT) to "disprove" the hypothesis that signal
processing occurs after the initial capture.

However, I believe that analysis was flawed because he examined a well exposed
image with a high Signal to Noise Ratio (SNR) which obscured the result.

I applied Emil's FFT technique to the lens cap shots used to prepare the
earlier post and verified that signal processing (noise reduction) __does__
occur.

Without going into the gory details, all we need to know is that if the pixels
are uncorrelated (have nothing to do with their neighbors) then you get a
uniform distribution.

But if the neighboring pixels are correlated (due to noise reduction,
sharpening, signal processing in general) then the distribution will not be
uniform.

In the case of Noise Reduction (NR), a brighter center with dark edges is
predicted.

Here are the results for the D300 after an aggressive Levels adjustment to make
things obvious:

The shutter times were:

1/15 |
1/8 |
1/4 |

1/2 |
1 |
2 |

4 |
8 |
15 |

And we can clearly see noise reduction has been applied at 1/4s and longer
exposures.