Please Help! Single Characters Recognition after OpenCV Preprocessing

58 views

Skip to first unread message

Arthur Cruz de Araújo

unread,

Sep 16, 2015, 4:26:33 AM9/16/15

to tesseract-ocr

Hello there,

I'm trying to implement OCR for license plate recognition. So far, I'm able to cut out each character individually using OpenCV and put it through Tesseract. I uploaded an image with my result after detection. About recognition, well, this is my first time using Tesseract so I'm simply selecting -psm 10 so each image is treated like a single character.

Most chars are being recognized correctly, but the "W", in 2.jpg, is returned as "H" and also the zero "0", in 6.jpg returned as "l" (low-case L). I suppose that the problem with the zero is more easy-going since I can assume that the last fours images contain numbers and Tesseract might get it right under this assumption, is that right? However, I'm really lost on the W problem. Maybe I could design some neural network to help me out, but I believe Tesseract can do this, I just don't know how yet.

Can anybody help? :)
Thanks in advance!

Arthur