General text detection / extraction problem

asked 2018-09-24 06:21:02 -0500

holger gravatar image

Hello,

I have trained a yolo model which can detect objects. Great. On that objects i have sometimes text i want to extract. To be precise: Text from numberplates.

I think OCR is what opencv offers for me for text extraction. The alternative would be to train a text classifier(maybe on top of existing) model

Does anyone have practical experience regarding: - speed of ocr - precision of ocr

Especially in comparison to a cnn approach. My head/assumtptions tell me that cnn approach can be better. Because the general pattern seems to be - let a network reognize the rules instead of trying some self written algorithms.

So any hints or practical tips / opinions would be highly welcome. Greetings, Holger

edit retag flag offensive close merge delete