|
|
Hi Tim,
Did you try noise filtering by a median filter? The median
should have a smaller radius than the variance.
If the noise is "well-behaved" (Gaussian, no hot pixel clusters),
and the features in the image have high contrast, you could
also try a median filter that preserves small objects with high
contrast, e.g. with the following macro:
threshold = 50; //noise must be lower than this
title = getTitle();
run("Duplicate...", "title=median");
run("Median...", "radius=4");
selectWindow(title);
run("Duplicate...", "title=selectiveMedian");
run("Remove Outliers...", "radius=4 threshold="+threshold+"
which=Bright");
run("Remove Outliers...", "radius=4 threshold="+threshold+"
which=Dark");
imageCalculator("Subtract create 32-bit", "median","selectiveMedian");
imageCalculator("Add create 32-bit", title,"Result of median");
run("Median...", "radius=0.5"); //hot pixels
selectWindow("median");
close();
selectWindow("selectiveMedian");
close();
selectWindow("Result of median");
close();
---
If you use larger magnification, you can also use a larger
radius for the variance filter, so the average over the
noise will vary less. This will make it slower, so an
alternative might be reducing the image size (Averaging
reducer on the ImageJ Web site), which will also reduce the
noise.
Michael
________________________________________________________________
On 25 Sep 2007, at 22:41, Tim Kline wrote:
> I have been using the variance filter to compare images with
> contrast differences in features greater than 3 pixels in size.
> However, noise is larger now that I work at higher magnification
> and variance does not pick up the difference between the images
> now. Does anyone have any suggestions for analysis? I have tried
> to use Gaussian blur and then variance to remove the noise with
> little success.
> Tim
|
Locked
