Handwritten OCR capabilities. How to train the engine

156 views
Skip to first unread message

cf.an...@gmail.com

unread,
Feb 24, 2013, 5:19:56 PM2/24/13
to ocr...@googlegroups.com
Hello all,

I have been to the painful job of training tesseract and learn about it just to be quite disappointed.

I am hoping that OCROPUS is easier.

First question I would like to ask is to learn about any kind of documentation on how to train it for handwritten recognition.

Has anyone tried it? I need a OCR engine that can do both handwritten and normal typewriter recognition from older forms.

Also, is there any image preparation that I need to go over before using the engine?

As I am new to it, I appreciate any pointers specially those related to handwritten recognition.

Thanks.

Raj Julha

unread,
Feb 25, 2013, 10:17:46 PM2/25/13
to ocr...@googlegroups.com
I tried historic handwritten documents with version 0.4.4 for my final year dissertation and failed. The main issue I had was the character breakdown, the tool didn't separate the characters correctly so I couldn't create a set of training data. I haven't tried the latest version  so cannot comment on that.

Good luck

Raj

Tom

unread,
Apr 10, 2013, 1:39:50 AM4/10/13
to ocr...@googlegroups.com
The latest version of OCRopus (0.7) has a new recognizer based on recurrent neural networks.

It should be easy to train for printed materials, but it probably isn't so good for handwritten inputs in the current release.

Tom
Reply all
Reply to author
Forward
0 new messages