As a hobbiest stamp collector, I often run into the need for classifying stamps based on minute differences, such as these:

varieties of the 1 pence Machin; courtesy of the Deegam catalogue

Now, I literally have thousands of them (in ziploc bags) and I am planning on building an automation platform, that makes a picture and immediately classifies the stamp as one of these varieties. (I make this type of integrated systems for work, so doing it at home is very feasible.)

What I am stuck at, is at identifying what type of image recognition / AI / classical fitting problem this is. It is somewhat made difficult, by the fact that the overall variation between stamps is quite large: cancels can obscure part of the image, there might be damage, color varieties, etc.

This is a representative image of what an algorithm would have to deal with:

enter image description here

I can imagine it to be a step wise approach:

  1. Normalization: crop, rotate and scale the stamp
  2. The hardest part, where any help is appreciated!


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Browse other questions tagged or ask your own question.