[PDFBOX-3248] Unwanted spaces in text extraction (2) - ASF JIRA

Voters

Watch issue

Watchers

Link

Clone

Update Comment Author

Replace String in Comment

Update Comment Visibility

Delete Comments

XML

Word

Printable

JSON

Details

Type: Bug
Status: Closed
Priority: Major
Resolution: Duplicate
Affects Version/s: 1.8.11, 2.0.0
Fix Version/s: None
Component/s: Text extraction
Labels:
- ActualText

Description

The attached file provided by Francisco from the user mailing list has spaces in text extraction regardless of setting spacingTolerance or averageCharTolerance. I was unable to extract "Cada frasco ampolla" which looked straightforward in rendering, but it always appeared as "Ca da fras co ampo lla". Adobe Reader has no such problem.

The content stream has this:

     6 0 1.058 6 122.0924 312.51 Tm
     (Ca) Tj
     /Span << /ActualText (\376\377\000\255) >> BDC
       ( ) Tj
     EMC
     [ (da ) -301 (fras) ] TJ
     /Span << /ActualText (\376\377\000\255) >> BDC
       ( ) Tj
     EMC
     [ (co ) -301 (ampo) ] TJ
     /Span << /ActualText (\376\377\000\255) >> BDC
       ( ) Tj
     EMC
     [ (lla ) -301 (con) ] TJ

So there are really spaces there, and we keep them. Adobe is smarter, and ignores them because they are overwritten thanks to the "-301" backwards positioning.

Would /ActualText help? However it is always the same here...

Would it help to ignore spaces and decide based on positions only, maybe as an option? I added these two lines below the first existing one:

                String characterValue = position.getUnicode();
                if (" ".equals(characterValue))
                    continue;

The output looks promising:

F ó r m u l a :
Cronopen® Balsámico Adultos:
Cada frasco ampolla contiene: ampicilina (como ampicilina sódica)
100 mg; ampicilina (como ampicilina benzatínica) 500 mg.
Cada ampolla solvente de 5 ml contiene: dipirona 1000 mg; guaife
nesina 100 mg. Exc.: bisulfito de sodio; agua destilada.

A complete test brings many differences, most are harmless or are improvements. Only one test case really fails, hello3.pdf. Original extract is "Hello محمد World.", new extract is "Hello .Worldمحمد".