Segmentation validation: Calculate relative area in super-object and relative position to super-object?
Hello,
I'm attempting to assess my segmentation results against a set of reference image objects that I have manually digitized. After performing an overlay function between the segmented image objects and the reference image objects, I would like to calculate the metrics relative area in super-object and relative position to super-object. I have read in published papers that these metrics can be implemented in eCognition, but I can't work out how to calculate these metrics using eCognition. Is there a particular tool or algorithm that calculates these metrics?
Many thanks,
Sarah
Was this article helpful?
3 comments
Hi Sarah,
there is no specific tool or algorithm in eCogmition for this task, but it is certainly possible to do it in eCognition. Have a look at the relational features to sub- super-objects this will help you to get relative (or absolute) measures between objects on different levels. One would be "area of sub-object" that came to my mind when I read your post. You probably will need to compute customized features / object variables based on existing features as well (sky is the limit ;)). If you have an rule set, please post it, would be interesting for me, but also probably for the Community.
Cheers,
Matthias
Thank you, Matthias.
I would also advise you not to get hung up on the image object that are a direct result of a segmentation algo., these are simply initial image objects and a starting point for your analysis. You really need to consider end result. How do the image objects compare after running the strong OBIA tools for object refinement etc.?