Scanned images not working very well

[harnit-bakshi]

Hi I ran the demo.py scripts on many images most were fine like invoices, natural images But when I ran a scanned image it was not able to detect text reliably at all

Please see below: Original image:

Text box detection:

Could anyone please let me know if I need to retrain the model to support this? I am guessing I would need to fine tune the model with the scanned images?

Any suggestions highly welcome

[harnit-bakshi]

Looks like this repo is not very active?

[Mohamed209]

i think this network is more suitable to scene text detection , so it performs well in scenes and invoices , but not for documents check official paper https://arxiv.org/abs/1609.03605