Identification of Segmentation parameters (PCA SVM)

Hi,
I need to identify suitable segmentation rules based only on RGB value to make the distinction between 5 area of my pictures.
I performed a statistical analysis (PCA) with each pixel value and I know that RGB channels are sufficient to have a good segmentation.Then thanks to a Support Vector Machine analysis i identified the discriminant functions to make the segmentation between my 5 classes.
However this process is relatively long and not intuitive for someone non familiar with statistics and i would like to use a more graphical approach such as the one in the plugin Weka where the user have to draw “class”.

My problem is that i cannot find a way to extract from the weka classifier its segmentation rules.
If it is not possible, is there a another plugin which helps to identify segmentation rules?
Thanks in advance
Alban

Hello @Alban_Favre and welcome to the ImageJ Forum![quote=“Alban_Favre, post:1, topic:5492”]
i cannot find a way to extract from the weka classifier its segmentation rules.
[/quote]

That depends on the classifier you choose. For example, if you use a simple decision tree (J48 in weka), you will see the decisions taken by the classifier in the Log window:


Have a look at the classifier options as well under weka.classifiers.rules:

Thanks that’s helped a lot, i’m woking now with SMO or J48.
Regards

1 Like