Posted by
Jeremy Adler-2 on
Oct 23, 2017; 10:47am
URL: http://imagej.273.s1.nabble.com/Eastimation-signal-to-noise-ratio-on-imageJ-tp5019580p5019586.html
I have always been a little suspicious of signal to noise - it clearly works for simple signals like morse code, where it is very clear which is the signal and which is noise, and importantly the signal is nominally constant (noise aside).
In most images the signal is variable across the image so (a) there is usually an overlap between the high end of the noise range and the low end of the signal + noise and also a set of edge pixels that include a mix of background and foreground , (b) the mean, while useful, can be deceptive especially if the noise comes from variation in the number of photons - a particular problem in confocal microscopy with a standard deviation being the square root of the longterm mean number of photons for each pixel - it therefore varies from pixel to pixel.
An alternative is to compare two replicate images, which are nominally identical (if the specimen is static) but differ due to noise. This can be displayed using a scatterplot - something it would be great to see in acquisition software to give a clear visual indication noise. We have used the correlation between replicate images as a measure of noise then used this to then correct colocalization measurements between confocal fluorescent images made by correlation.
Replicate-based noise corrected correlation for accurate measurements of colocalization
Journal of Microscopy
Volume 230, Issue 1, April 2008, Pages: 121–133, J. ADLER, S. N. PAGAKIS and I. PARMRYD
For binning - you can bin your own image
Image / Transform / Bin
Jeremy Adler
-----Original Message-----
From: ImageJ Interest Group [mailto:
[hidden email]] On Behalf Of thpnick688
Sent: 22 October 2017 17:18
To:
[hidden email]
Subject: Eastimation signal to noise ratio on imageJ
Hi, everyone,
I have encountered a question on estimating signal to noise ratio on imageJ. I have an image with 4x4 binning. The binning is done by the CCD detector not by the software. I use the ROI to measure the mean gray level intensity on the main object in the image. I also use the same size ROI on the background to measure the background noise. I use the simple equation "S/N= “mean gray level intensity in subject” / “mean gray level intensity of the noise”" to estimate the signal to noise ratio. However, by comparing with an image that is unbinned, I didn't see the SNR is improved after the calculation. So, what mistakes did I make?
RL
--
Sent from:
http://imagej.1557.x6.nabble.com/--
ImageJ mailing list:
http://imagej.nih.gov/ij/list.html--
ImageJ mailing list:
http://imagej.nih.gov/ij/list.html