I want to make an mobile app where you scan using the camera and it gives you the card you just scanned. (from a board game)
I got a PNG of each and every single existing cards, but I don't really know what to use.

For now, I've been using the text in the card using OCR to find the card, but it's not 100% accurate as the typography is pretty special.

I've been thinking about using image classification, but I'm not sure it's the right way to do it, as I would be training for hundreds of classes. (got over 400+ different cards)

Do you have any idea what I should be using ? (I'm sorry if this seems like a basic question, but I just started learning about CV and ANN)
Thanks for your help !


First Approach:

You can probably train a CNN ( Convolutional Neural Network ) on the images you have. If the number of images is less, I prefer you to use Transfer Learning.

In Transfer learning, we have a highly trained model ( like Inception, ResNet ) and we modify it so that it can classify images which we need.

Second Approach:

Use object detection. You need to localize an object ( card ) from the image and then classify that cropped image. This could be tedious if you're a beginner.

You can refer to these links:

You need to train a localization model which can track and crop your images ( of cards ).


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.