mittagessen / kraken

OCR engine for all the languages
http://kraken.re
Apache License 2.0
708 stars 129 forks source link

polygonisation for vertical writing systems #625

Closed dstoekl closed 6 days ago

dstoekl commented 1 month ago

this was better in the past I believe. Lines were manually drawn and then subsequently drawn by kraken in eScr. image

https://msia.escriptorium.fr/media/users/2/export_doc4710_trial_07_24_alto_202407181035.zip

colibrisson commented 1 month ago

Are you using center-line or baseline?

dstoekl commented 1 month ago

center-line !

dstoekl commented 1 month ago

Here is another very bad polygonisation: data: image model: https://msia.escriptorium.fr/media/models/187ba3fc/jap_seg_no_aug_1.mlmodel

https://msia.escriptorium.fr/media/users/2/export_doc4710_trial_07_24_alto_202407190838.zip

colibrisson commented 1 month ago

I always get weird polygons with center-line. I have tried to fix this issue multiple times.

It is odd thought because at the end center-line option don't do much: it only impact the offset of the baseline.

dstoekl commented 1 month ago

do the polygons always look as if the wind blows from the West (left)?

colibrisson commented 1 month ago

I think so.

dstoekl commented 1 month ago

west storm: image

mittagessen commented 6 days ago

The general fix for the polygonisation seems to address this as well.