Highlighting borders on images

Question

Image processing

What are the criteria for assessing the quality of contouring algorithms (boundary detection)?
Is it possible to somehow evaluate the noise resistance of edge detection algorithms?
Please provide literature.

Answer 1

I doubt that it is possible to formulate a universal mathematical criterion. For each task there will be its own, empirical. And if the criterion is empirical, then the only way to evaluate is the reference images.
I would advise you to do this: write an algorithm that generates pictures that are approximately similar to those that need to be processed. At the same time, let it save the reference shape of the borders for each generated image. Then fills them with noise and other debris. Then you start up your selection algorithms and see how the resulting image matches the reference one.
When we set up one algorithm at one time, we processed a large database manually, marking all the boundaries that the algorithm was supposed to catch (search for eyes). And then they ran the algorithm over the entire database and looked at how much where it did it, evaluated its effectiveness.

Answer 2

Comparison of the work of the algorithm with the reference selection. Then demonstrate the percentage of "hits" and "misses"

Answer 3

" Canny edge detection enhancement by scale multiplication ", first page and links to it.