TanguyJeanneau
commented
1 month ago Hi @EliteWeapon !
This is a good point, iris recognition systems do need a form of image sharpness verification to ensure that it makes sense to process the image.
On the Orb, we have a dedicated model that takes care of that. We don't have clarity on the timeline for its open-sourcing, so I cannot commit to anything here.
However it would be super useful for the community to have a form of image sharpness verification, regardless of this other model. I can see plenty of ways to do so: Mean image gradient magnitude, Convolution-based methods (see appendix of John Daugmann's paper), Contrast estimation, etc.
@EliteWeapon are you interested in contributing ? You seem to already be familiar with open-iris, this would be impactful for the project and I think the method in the paper above is rather straightforward.
Is there a plan to validate the clarity of iris?
Context
Expected Behavior
The algorithm can validate the clarity of iris or ir-image.
Current Behavior
For unclear iris images, the algorithm can also work, but the final codes are not valuable. Such as the following image:
The result of this image processed by the algorithm is: Geometries:
Normalized iris:
Final iris template:
