Binarization: wrong concept?

[tischi]

@manerotoni @stemarcotti

Currently, in the activities, the binarization module teaches how to create a binary image by means of a simple intensity threshold. But the concept map (and example figure) teaches the general concept of converting an intensity image to a binary image by whatever means (could also be a DCNN). So there is some mismatch...

I somehow feel that "binarization" might not be the right concept. I am not sure how to do it better, but some thoughts:

• Maybe the concept of a binary image should become part of the data types module (and its importance should be stressed more there)?
• Currently we are lacking the concept of "semantic segmentation". One could say a concept is: intensity image => semantic segmentation, and if there are only two classes (foreground and background) then the datatype of the semantic segmentation image happens to be a binary image.
• I just googled and "foreground background segmentation" also seems to be a thing, e.g. see here, where, interestingly they do not use a simple intensity threshold.

My current feeling is that "Binarization" is not the ideal concept to teach, better would be "Semantic segmentation" and "Intensity threshold" (more ideas for intensity threshold).

Maybe we just add a module "Semantic segmentation" that contains no activities, but just serves to explain the concept (probably including a label mask image as a storage mechanism for the segmentation)? Then a module "Foreground background segmentation by an intensity threshold" or just "Intensity threshold" (?) and there we refer to "Semantic segmentation" and say this is the most simple version for of it?

Maybe something totally different? What do you think?

[tischi]

I think I have an idea 😸

What about we create two new modules:

• Semantic segmentation
• Instance segmentation

And the Activities and Exercises in these modules are not about how to create those segmentations, but how such segmentations are typically stored (label mask images) and how to inspect already existing segmentations**; stuff like: how many objects are there, how many semantic classes are there, are they disjoint, what class covers the largest part of the image, ...

I think this could be really good, because it enables the students to understand the output of all the bizillion methods and tools that segment stuff in images.

[tischi]

Maybe better: just one module: "Image segmentation" and there we explain the difference between instance and semantic and that both can be stored as label masks.

[tischi]

=> rename to title: Thresholding tags: Foreground background semantic segmentation