In Operation
Surya predicts line-level bboxes, while Tesseract and others predict word-level or character-level. Let’s test the software with the nyt.png image file from their GitHub repository.
It writes a JSON with the detected bboxes, and saves images of the pages with the bboxes if you include the --images option.
Here’s a generated image showing the bboxes.
Summary
Surya has just seen its first public release but it already looks promising.
We’ll look at the software in more detail once it has added text recognition and table and chart detection. We’ll then evaluate the software for inclusion in our OCR Systems roundup.
Website: github.com/VikParuchuri/surya
Support:
Developer: Vik Paruchuri
License: GNU General Public License v3.0
Surya is written in Python. Learn Python with our recommended free books and free tutorials.
Pages in this article:
Page 1 – Introduction and Installation
Page 2 – In Operation and Summary
Given the weights are not open source, is this really compatible with an open source license? I’m not a legal expert.
That’s a good question. There are lots of open source projects that have CC elements. You see that all the time in open source games.
This really needs text recognition. When is that going to be implemented?
Soon.