Select Region Based on Texture

Hi all, thanks for thinking on this query.

I was wondering if I could select a region of my images based on it’s texture. I have attached an image that exemplifies the different potential regions that I will need to distinguish.

Thanks
Neil


Hi Neil,

If you have a Windows machine, I would suggest using the ilastik software package that comes bundled with CellProfiler. You would train a classifier with Ilastik and then saves the classifier. Then, you uses the ClassifyPixels module to classify the pixels in an image. ClassifyPixels produces an “image” consisting of probabilities that the pixel belongs to the chosen class; this image is similar to an intensity image that would be produced by fluorescence imaging. Provided that the classifier is sufficiently accurate, the image is well-suited for input into one of the Identify modules for object detection.

Detailed instructions on using the ilastik interface can be found here.

Regards,
-Mark

Yes! Mark beat me to the description, but I just tried this quickly and this is what it looks like in ilastik: cl.ly/image/1M2a1h3v1A26
I am attaching the ilastik classifier for this single image based on the training set shown in the screenshot (unzip it first).

And then in CP you get output that looks like this: cl.ly/image/391j3m0a2s2l
The version 11710 pipeline is attached.

Good luck!
David
ilastik_classifier_output.h5.zip (311 KB)
pipeline_classify_ilastik.cp (2.49 KB)