View Single Post
  #2  
Old 29-06-2010, 11:51 PM
luigi
Registered User

luigi is offline
 
Join Date: Jun 2010
Posts: 438
There were some incorrect comments on the former thread so I will continue here.

First of all two keys:

To minimize noise you always want to expose longer, when you can't expose longer if the histogram still allows it then increase the ISO. Higher ISOs have better signal to noise ratio than lower ones.

A 1 second exposure at ISO800 has LESS noise than a 1 second exposure at ISO100.

ISO is a hardware amplification of the signal that the camera sensor reads. We can divide sources of noise in two categories:

1) Noise that is amplified by the ISO gain (Npre)
This includes:
- Photon fluctuation noise
- Read noise (sensel read error)
- Thermal Noise

2) Noise that occurs after the ISO amplification (Npost)
This includes
- Analog to digital conversion error
- Rounding to fixed point

So we can formulate the signal as:

[(So +/- Npre) * ISO] +/- Npost

In this formula we can see that Npost is constant and doesn't depend on the ISO amplification therefore as the ISO gets higher the Signal to Noise Ratio improves.

It is true that as exposures get longer Npre starts to dominate over Npost but it is also true that a higher ISO can't have a worst S2N ratio than a lower one so you should also use the highest ISO you can afford for the longest exposure you can afford.

This holds only for native ISOs, ISOs that use hardware amplification. In my camera those are ISO 100,200,400,800,1600 and 3200. Intermediate ISOs and expansions (fake) are not useful because they use just a software amplification and therefore amplify both Npre & Npost noise.


This paper from Dr Emil Martinec is a good reference:
http://theory.uchicago.edu/~ejm/pix/20d/tests/noise/

This graph shows it nicely:
http://theory.uchicago.edu/~ejm/pix/...Rwindow1d3.png

Luis
Reply With Quote