sanitize: stay on page image/array

[bertsky]

Fixes #21 – it was not correct to use the region image/array here, because that depends on the bounding box of the region, which can be too small.

Something not covered by this is when TextLine coordinates even extrude the page Border.

[wrznr]

I get it!

[bertsky]

One more thing about this fix. It is not optimal w.r.t. rotation.

If deskewing has happened (only) on the region level, then the closing of the regions' lines will no longer find deskewed hulls, but undeskewed hulls. That's because of the simplistic approximation to morphological operations used here (simple rectangular minimum and maximum filters) for efficiency reasons.

[wrznr]

But deskewing on region level would usually happen after sanitizing the regions, right?

[bertsky]

But deskewing on region level would usually happen after sanitizing the regions, right?

I don't think so. Sanitizing regions is only possible after textline segmentation. But the latter usually relies on deskewing itself.

[wrznr]

Right. In my particular usecase (reusing existing ABBYY segementation) however, I apply region-level deskewing after sanitization (results look great!).