Document layout analysis - superscript / subscript

[jamesanastasi]

I'm having a bit of difficulty with this particular use case :

When a line has superscript the line extraction tends to extract the superscript word as a new line. this is bothersome because the word ends up in the wrong place in the raw text. exemple : from the example PDF

TestPDF5.pdf

Integer egestas tristique ^aliquet. Sed consequat massa non vehicula _finibus

is interpreted B1 : aliquet. Sed consequat massa non vehicula finibus B2: Integer egestas tristique

So the raw text is :

aliquet. Sed consequat massa non vehicula finibus Integer egestas tristique

I have adjusted the DocstrumBoundingBoxes parameters : BetweenLineMultiplier to .75 and I get the words in the right order
B1 : Integer egestas tristique B2 : aliquet. Sed consequat massa non vehicula finibus

but this creates a new probleme :

The two blocs at the end : where each a bloc has two lines ..

Sed a felis fringilla,                           Praesent elementum in enim maximus libero sit amet.                  id sagittis.

After changing the parameters to make the superscript they are split up into séperate blocs ( and therefore loose their order)

B1 : Sed a felis fringilla, B2 : Praesent elementum in enim B3 : maximus libero sit amet. B4 : id sagittis.

I've tried different variations of recursive XYCut and played with the ordred blocs but can't seem to find the softspot where I get the blocs and the right order.

Any suggestions or ideas would be appreciated

[davebrokit]

Feels you need to use the reading order detectors:

https://github.com/UglyToad/PdfPig/wiki/Document-Layout-Analysis#reading-order-detectors

You might need a higher T parameter to the reading order detector than 5.

If that doesn't work out of the box look at the code in UnsupervisedReadingOrderDetector.cs and IntervalRelationsHelper.cs. You can use that to create your own reading order detector where you use the relationship between a superscript box and a normal text box to impose a reading order.

If you get it working it might be worth raising a PR with an update to UnsupervisedReadingOrderDetector :)