Ask HN: What's the best out-of-box Document OCR/Analyzing/recognition API?

7 points·pjerryhu··9 comments

I tried with Google Document AI and AWS Textract, they both seem a bit hard to use and AWS is especially pricy

"Show HN: BetterOCR combines and corrects multiple OCR engines with an LLM" (2023) https://news.ycombinator.com/context?id=38056243 :

junhoyeo/BetterOCR: https://github.com/junhoyeo/BetterOCR :

> Better text detection by combining multiple OCR engines (EasyOCR, Tesseract, and Pororo)

Thank you so much! We went with AWS finally because we have startup credits, but good to know the options and previous discussions here!

Unstract, they are a month away from open source launch of automating document based workflows. They also have plug and play APIs Link - https://unstract.com/api-for-unstructured-data/

Thank you!

I was using a Python module called easyocr. It was working for a while, but stopped recently.

I switched over to pytesseract, a wrapper for tesseract. It was pretty simple to use and I got it working in 10 minutes.

How it stopped? it was a performance problem?

No idea why it stopped. I tried reinstalling it, but it is not working with simple examples.

Tesseract is working so I am sticking with that for now.

tesseract OCR is open source and very good. https://github.com/tesseract-ocr/tesseract

Tesseract can run locally and you can also use other open source OCR models which use neural networks. I am pretty sure it is possible to fine-tune open source LLMs on images and get reasonable results.

Reply on news.ycombinator.com