I am dealing with hierarchical data for mapping some elements under the sea from orthophoto imagery at 0.5m ground pixel resolution. What I would need is to be able to easily define which segmentation level is above which. I do not use multi-resolution segmentation for this since it eats up too much processing time and I use contrast split instead since it loops and extracts specific classes automatically (runs on a layer obtained from unsupervised classification to pull out, as an object, all of the obtained values without the need to manually enter each of those - THX Mathias for help here). 

The next step is to have segmentation above it where I managed to figure out how to classify above water objects [coast, rocks, boats through the use of abs. mean deviation filter (prototype)] and form a super object that I need to use then to remove the classification from all the sub-objects that are overlapping with this one big object above).

Now, the problem here is to be able to easily control which layer is above which. I am wondering, is it possible to generate an algorithm that will allow me to assign hierarchy values to levels easily? So I can say - the coast segmentation level is 1., the unsupervised thresholding level is 2. and similar (vertical hierarchy). This way I can rearrange them also at a later point if necessary. 

Another one that popped in my mind, can I automatically see what is the percentage of each under level class within my above level class without generating such a feature for each class and then checking it out?

That would make things easier since I am trying to automate a process that generates around 30-50 classes over each iteration and then tells me which classes are within and in which percentage of the above-generated chessboard segmentation. Would be cool to be able to count amount of specific classes within one above segment and use that too. Great for habitat complexity automation. 

Best regards to all! :)


  • Matthias Staengel

    Hi Ivan,

    I will just write down what comes to my mind ;).

    Regarding the image object Level hierarchy. Okay... I tried to write something but then I noticed, that I probably don't understand what you want to do ;). In general you can have numerous image object levels in a map: "The levels of an image object hierarchy range from a fine resolution of image objects on the lowest level to the coarse resolution on the highest. On its superlevel, every image object has only one image object, the superobject. On the other hand, an image object may have – but is not required to have – multiple sub-objects."

    You can use "copy image object level" and choose "above" or "below" and this will insert a copy of the defined image object level above or below. I am happy to brainstorm more, but I think I do not get the set-up.

    For the percentage of sub level classes you can use the majority vote feature (we have area and count). A little tip, if you have created the feature "rel. area of sub-object" for one class, you can right-click on it and choose "create for all classes". This will automatically create this feature for all classes!

    Don't know if any of this helps, at least I tried ;).



  • Ivan Tomljenović

    Hi Matthias!

    "You can use "copy image object level" and choose "above" or "below" and this will insert a copy of the defined image object level above or below. I am happy to brainstorm more, but I think I do not get the set-up." - THIS is what I needed. Thank you! :D From the eCognitionTV video on hierarchy, it looked like only multiresolution segmentation allowed you to choose the position of the layer. :) Now it's clear! 

    And a great tip for the class selection! thank you!




Recent Discussions

Need more help?

Enter Knowledge Base Contact Support