Textlines detection algorithm used in tesseract

[张晓艺]

Hi all,

I'm looking a non-deeplearning-based textlines detection/segmentation algorithm, since my problem is relatively easy (without complexed background). I wrote several scripts with a combination of OpenCV ops like dilate, erode and findContours etc.. But I found it's very sensitive to hyperparam changes (like iterations of dilate). 

I know Tesseract're using a non-deeplearning-based textlines detection (as a part of Page Segmentation) and believe many efforts have been put on tuning it. Did you guys have any idea about the paper/code I could refer to?

Thanks,

Xiaoyi