Hi!

Here you can find a master thesis about all evaluation indices for semantic segmentation. Hope it helps :)

Thesis Segmentation Validation Framework

From my experience, it depends on your application. Some time HSI is a good choice. In other cases you may use r, g, b, X, Y, Z. It depends on your application.

Dear Donatella Giuliani,

As far as I know, it depends on your purpose and application. Although there are several algorithms for benchmarking of image segmentation neither directly mention the suitable indices for segmentation. Hope the answer of your question will be in the following links.

Conference Paper Near set index in an objective image segmentation evaluation framework

www3.ntu.edu.sg/home/asjfcai/ICME13_Hui_benchmark.pdf

Dear Donatella,

I have succesfully used Hue angle in an HCL colour space derive from CIEL*a*b* to sparate and re-colour individual objects in a complex coloured image. See my RG contribution 'Computer Systems for the Specification, Visualization and Communication of Colour'

Note however that I was using accurately calibrated on-screen colour at the time, in effect, the better the calibration the better the object isolation.

Regards and best wishes

David

Thank so mauch for your accurate answer

Donatella