I’m looking to start a small python project which heavily involves OCR. I would like feedback from others on which OCR tech is easier, but more customizable to use. This the list of OCR tech I’m looking at pytesseract, EasyOCR, Keras-OCR, and Amazon Textract.
I was using tesseract once. Even with the right locale, it would, among other things, regularly confuse points and commas within numbers. While this was a dealbreaker for me, I have to admit that I did not spend too much time experimenting with it further. Instead, I was lucky to be able to just extract raw text from the PDF files.
So this is rather anecdotal.
maybe this project’s output / results can help with your research: - OCR-D
they heavily employ Python in their stack. the stack itself is suited for larger institutions.
as a sidenote: the term OCR is technically obsolete as most solutions are based on neural networks that operate on the line level (among other reasons to facilitate the transcription hand-written text). the discourse tends to call it Automated Text Recognition (ATR) or simply Automated Transcription nowadays. (of course some will throw some AI hype into their terminology too.)