Train pixel classifier (classification/probability)

Dear all,

I have a question regarding the use of the pixel classifier.

I use this function to measure the surface area.

What is the difference between the function Output: classification/probability. When I use these two functions, I observed a difference between the detected tissue; setting classification looks more dense then when I use probability. Probability has more ‘shadows’. However, by using both functions the surface area measured was the same.

And is there a possible way to distinguish two different classifiers for instance: is it possible to turn off one of the two classes you have set? I like to know whether a positive signal (red) is included in the measured tissue area (yellow)?

For example I have a picture below:

Thank you!


If you click Show, it should output a multichannel file showing each individual class. The probability maps might be easier to read then.

1 Like

Hi @K.massy in general, the end results should be the same whether you use classification or probability for the output.

Basically, classification is the best one to use most of the time – it makes clearer how exactly each pixel will be interpreted, and it uses less memory. Probability is useful as an output if you also want a measurement of how confident the prediction is (it’s not a real probability, but it’s as close as we can get…). Here, the intensity of the color relates to that confidence.

In the end, the pixel classifiers produce a single ‘winning’ prediction for each pixel and that should be the same, regardless of which output option you choose.

Thanks @Research_Associate for your reply. I can indeed visualize the separate individual classes!

1 Like

Hi @petebankhead! Thanks for your explanation. Now it is clear to me how to use these two settings for the pixel classifier.