I would like to evaluate an idea for OCR of handwritten texts (especially with mathematical contents).

Is there a dataset for OCR of handwritten texts available? It should contain images of documents, e.g. something like

enter image description here

It would be good (but not necessary) if the dataset contained information about the language(s) of the document and probably also the topic.

Side questions

I'm not sure how OCR algorithms are evaluated, as OCR contains multiple problems:

  1. Segmentation: Find single entities in the source image
  2. Classification: Classify the segmented images
  3. Placement: Find the relationship between two classified symbols (is one to the right of the other or rather the the right-top? Hence: Is it ax or a^x?

I'm also not sure how such a dataset should be labeled. But for this question, I would also accept datasets which do not contain labels.

  • Probably icdar2011competition. I've just registered, but I'm waiting for an email right now. Feb 10, 2017 at 13:05
  • Have you had a look at this ? Feb 13, 2017 at 17:35
  • @ettore this is an online recognition dataset. Ocr is offline recognition Feb 13, 2017 at 18:34
  • 1
    Yes, but i've seen here that the inkml dataset can be converted to png. I thought that it could be usefull for you (Another well known database of handwriting texts, by the way). Feb 13, 2017 at 18:59


Your Answer

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

Browse other questions tagged or ask your own question.