Making a binary skeletonization less noisy - image processing

Hi - I am trying to skeletonize a binary mask like this https://dl.dropbox.com/u/54307333/4ANDmask.bmp. But I end up with too many small branches coming off of the long segments https://dl.dropbox.com/u/54307333/4Maskthin.bmp. I want to get rid of small branches (<20 pixels) that are only connected on one end. Ones that are connected on both ends are valuable to me.

I am using bwmorph to calculate the endpoints. And using that I can exclude branches that are below a certain size, but like I said, I want to keep the ones that are connected on both ends.

I probably have to do some kind of if statement, checking the connectivity of each endpoint...? Not sure. Any ideas?

Thanks,

Nad

Answer 1

[Image Analyst]

Try blurring the image before you skeletonize it.

[Nad]

You mean blurring the binary mask? I hadn't thought of that. I was blurring the original image before getting the binary mask, but not the mask itself.

Will try that.

[Nad]

For this kind of image, what do you think is the best way to blur? Simple median filter or should I go with something else?

[Nad]

tried gaussian filter (see help for fspecial if anyone is interested) with a decent sized window (15x15 for a 250x250 roi) and that got rid of the small off shooting branches. thanks!

also tried bwmorph(image,'spur',numofpixels) to get rid of branches of a certain size and that seemed to help as well.

[Image Analyst]

Just try a simple box filter on the grayscale image.

out = conv2(in, ones(21), 'same');

Then do your morphology.