headlogo2

The "Dark Noise" project

This page belongs to a much larger site that can be accessed through this link. However, it is mainly written in french.
I try to answer questions both in french and english.

Introduction

The goal of the "Dark Noise" project is to create a free database describing the behaviour of different camera sensors regarding black noise.
Most caracteristics of sensors are impossible to meter for casual users, however black noise can be analyzed without any special equipment.
The methodology is simple: a few shots have to be taken at different iso settings and different shutter speeds, with the lens cap on.
It is then possible to post-process the result to describe the behaviour of the captor regarding black noise.

How to

Simple method

Use a formatted memory card. It is not mandatory, but it's safer
Use only the camera RAW mode
Put the camera in manual mode.
Make a completely overexposed first shot.
Place the cover over the lens. Check that no light can enter through the viewer. Then make different series of shots. For example you can shot:
- 100 ISO: 1/1000, 1/100, 1/10, 1s et 10s
- 200 ISO: 1/1000, 1/100, 1/10, 1s et 10s
- 400 ISO: 1/1000, 1/100, 1/10, 1s et 10s
- 800 ISO: 1/1000, 1/100, 1/10, 1s et 10s
- 1600 ISO: 1/1000, 1/100, 1/10, 1s et 10s
You can also use different speed, different ISO settings, or shoot many more pictures. All of them will be used.
Put your RAW files in some place where I can get them. If you don't have any, send me a mail at sensor@photo-lovers.org and I will give you an anonymous ftp access to one of my servers.
Send me a mail at sensor@photo-lovers.org to tell me where the shots are located.

If you don't have a way to make the RAW files available

Use a formatted memory card. It is not mandatory, but it's safer
Use only the camera RAW mode
Put the camera in manual mode.
Make a completely overexposed first shot. Put the resulting file in the WHITE directory, and erase it from the memory card. Be careful to put only the RAW file in the WHITE repertory. Some cameras generate a second file, usually ending by JPG or THM, which must be deleted.
Place the cover over the lens. Check that no light can enter through the viewer. Then make different series of shots. For example you can shot:
- 100 ISO: 1/1000, 1/100, 1/10, 1s et 10s
- 200 ISO: 1/1000, 1/100, 1/10, 1s et 10s
- 400 ISO: 1/1000, 1/100, 1/10, 1s et 10s
- 800 ISO: 1/1000, 1/100, 1/10, 1s et 10s
- 1600 ISO: 1/1000, 1/100, 1/10, 1s et 10s
You can also use different speed, different ISO settings, or shoot many more pictures. All of them will be used.
Put all these files in the RAW directory. Make sure that no other file is copied in this directory (and especially not the overexposed shot taken above)
Click twice on getdata (or execute the program in a shell window). The program will ask a few questions:
- Owner name: you don't have to answer, or you can only provide a pseudo
- Serial number: Not mandatory, but will help in locating your camera inside the database
- Temperature (integer only): to take into account noise variation. Provide an integer value only.
After a few seconds or a few minutes depending on your computer and on the number of shots,getdata will generate a results.gz file. Please send this file by mail as an attached document to sensor@photo-lovers.org. The file is small and won't waste bandwidth.
After a few hours or a few days depending on my disponibility, the statistics of your sensor will be available from this page.
If there is a long exposure noise reduction mode, please repeat the above process for both modes (with and without noise reduction), or at least leave a note in your mail telling which was the mode used.
Caveat: this program uses the excellent David Coffin's dcraw program. Only cameras supported by dcraw are supported. The list of supported cameras is available on dcraw site.
Caveat bis: this program most certainly won't work with camera using Foveon sensors. If you have such a camera please send me a RAW file, and I might be able to support them.

There are different versions available for download:

The windows 32 bits executable should run on any flavour of windows XP (even 64 bits) and probably on older windows. Vista compatibility is not guaranteed.

There is no mac executable. If someone familiar with ocaml programming and with access to a mac wants to port the program, he can contact me through the forum.

Cameras currently available

How to read the results

Below, the results of some measurements for an APS-C sensor are displayed, at ISO 200 and for four different speeds (1/1000, 1/100, 1/10 et 1s).

ISO200

The best solution to understand how to read these data is to explain them on an example.

Let's suppose we have a camera with an 8MPix sensor, with a linear dynamic of 12 bits (values read range from 0 to 2¹²=4096).
If the RAW file contains 4100 times the value 256, we will plot the following point: on the x-axis, its position will be log₂(256/2¹²)=-4, and on the y-axis it will be log₂(4100/8000000)=-11

On the above curve, we see that most of the values are around -6, which represents, 1/2⁶ soit 1/64 of the maximum possible value, which is on a 12 bits sensors 2¹². So, the value -6 represents in fact 2¹²/2⁶=64 for this 12 bits sensors. Moreover, we see that this value has been read for 2^-3=1/8 of the total numbers of pixels; for an 8 MPix sensors, this is 1 million pixels.

This representation should help comparing different sensors, but unfortunately it is not as simple as that. All sensors are not calibrated in the same way. On the above example, the sensor used is in an EPSON camera (the RD-1), and the zero of the sensor is calibrated at 64. Canon sensors are usually calibrated around 128 or 256. It might seem strange that camera manufacturers don't calibrate their sensor at zero, because some of the dynamic is lost in the process. However, much more information is available regarding noise, as we have both part of the noise (above and under zero).

It is thus necessary to generate a renormalized curve for Canon (and other) cameras to have curves that can be compared. For this kind of camera, a figure labelled "ISO_n" or "SHUTEER_n" is available, with normalized values.

ISO200

It is not however enough to perfectly compare sensors and amplificator circuitry. Some manufacturers have implemented noise reduction systems, and the values stored in the RAW files have already been processed. This is a fact which is well known by astro-photographers, who are testing regularly cameras. For example, Nikon cameras (even the D3) are useless for astro-photography, as information from the sensor is already processed before being stored in the RAW file.

With many cameras, the noise reduction system (usually called long exposure noise reduction system) can be disabled by a menu. It usually substract a dark frame taken with the shutter closed to the original frame. Hot pixels and other sensor defects are then usually removed. However different cameras have different behaviour.

This is a comparison of shots taken with a Canon 40D with a shutter speed of ten seconds. On the left, NR is off and ON on the right. The results are perfectly in accordance with theory.

However, things are not so clear for the Canon 450D (below). The noisiest pixels are removed (you don't see any unpleasant things between -2 and 0) but on the opposite it looks like around -8 the NR system degrades performance. This could be explained if the 450D was already using a kind of NR system even when it is officially disabled...

The "High ISO noise reduction option" is completely pointless when shooting RAW. This is only a software noise reduction system used when doing in-camera JPEGs.

Homepage

The download and use of documents or photographies from this site is allowed only if their provenance is explicitly stated , and if they are only used for non profit, educational or research activities.
All rights reserved.

Last modification: 07:03, 10/27/2010 xhtml validation