# Measurement of image noise

Discussion in 'General Electronics' started by Sebastian, Oct 30, 2003.

1. ### SebastianGuest

Hello,

I´m working in a medical technology company and I have the following
problem:
We are looking for a new camera for a optical device. The amount of
light received by the device is poor, so the camera needs a high
amplification. Unfortunately this causes image noise.
I would like to compare different cameras and therefore I would like
to measure the noise. Is there a simple way the determine the noise
with a PC? I can grab a picture from the camera and do some image
processing on it. Is it possible to measure the noise with LabView or
the IMAQ Vision Builder from NI??? There´s no need for a precise
measurement I just want to compare some cameras. Thanks for your help!

Best Regards,

Sebastain

2. ### Hiep Quang LuongGuest

Just calculate the SNR (signal to noise ratio) A search on google gives:
http://searchnetworking.techtarget.com/sDefinition/0,,sid7_gci213018,00.html

Hiep

3. ### SteveGuest

The difficultly here being how to quantify the signal and the noise
power independently given the single measured signal.

Some useful measures i've seen are:

Peak-sig-to-noise (PSNR) => 20 * log10( 255 / variance )
this is based on 8bit data.

I've also seen => 20 * log10 ( ( 255 * 255 ) / variance ) used.
variance here is the global variance of the image [ie. second
statistical moment]

A measure for plain SNR is: 20 * log10( mean / variance )
again "mean" here is the global average.

teh scaling of the log function is less important if you're doing a
comparitive test between camera, as long as it is consistent.

As for how well each of these compare to a more informed SNR
calculation method, i don't know. [Anyone??]

Good Luck
Steve

4. ### AGGuest

Grab a copy of the IGOR Pro demo
<http://www.wavemetrics.com/Products/IGORPro/Demo.html>

You can use the image operations in IGOR to compute any measure you want
for the noise level.

The question of what would be considered a proper measure of noise is a
whole other issue. I would recommend that you pay close attention to
making sure that your system (sans camera) has stable/uniform
illumination. Ideally, to compare the cameras for a given light level
sphere or a very good diffuser). Once you acquire images of uniform
frames at the required illumination level you can compare the signals,
signal to noise and rms fluctuations as a function of position.

HTH,

A.G.
WaveMetrics, Inc.

5. ### mgGuest

The question is, how does he separate signal from noise?

IMAQ should have FFT routines. I would take a picture with no high spatial
frequencies (smooth, large features). After FFT, the high frequency can
then be attributed to noise.