Object Classification of US Quarter

I have seen many similar topics but none quite like this request.

I’d like to create a model that will categorize a quarter by year. Either the quarter will be for 1995 or 1998. If the quarter is 1993 then the quarter would not be classified.

I’m guessing I would have to set some confidence level in order to do this.

My actual image dataset would be larger but here is an example of how it would look.

1995 - 1995 — ImgBB
1998 - 1998 — ImgBB

What are your thoughts on the best way to proceed here?

before you can classify your coins, you’ll have to deal with the rotation
(e.g make them all point “upright”).

maybe you can obtain a homography matrix from féature matching
to warp them into the same pose.

then you can crop out the number part, and apply some machine learning (SVM/HOG, a small CNN, whatever)

1 Like

Very good idea! Thanks!!