[
https://issues.apache.org/jira/browse/PDFBOX-5868?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17878095#comment-17878095
]
Manish S N commented on PDFBOX-5868:
------------------------------------
the data is right there in the fonts, within reach
...
on second thought, yes there are problems.
this approach assumes that every glyph that has a corresponding unicode
character is _present_ is the cmap which isn't always true.
when seeing ligature data for நூ, we see
!image-2024-08-30-17-55-41-423.png!
here we can see glyph92 instead of the unicode character for dependent vowel ூ
which is not present in cmap because that actual glyph is never used in the pdf
(all pure tamil ligatures of ூ are irrugular and have their own glyphs and not
combined side by side unlike other dependent vowel glyphs so no use for the
actual glyph. hence pdf optimizers will chuck it away along with its unicode
mapping)
the problem is replicated by all ligatures of ூ (dependent vowel uu)
it is the case in languages like tamil but most other non-latin languages can
be fine. like hindi; it is a more regular language than tamil (when comes to
letters)
In the end there are also other problems like mangled cmap as a method of
obfuscation
> PDFBox not extracting text of non-latin languages(tamil, bengali) properly
> but adobe reader's save as text does
> ---------------------------------------------------------------------------------------------------------------
>
> Key: PDFBOX-5868
> URL: https://issues.apache.org/jira/browse/PDFBOX-5868
> Project: PDFBox
> Issue Type: Bug
> Components: Text extraction
> Affects Versions: 2.0.32, 3.0.3 PDFBox
> Environment: Ubuntu 22.04.4 LTS x86_64
> Reporter: Manish S N
> Assignee: Tilman Hausherr
> Priority: Major
> Labels: ActualText
> Fix For: 2.0.33, 3.0.4 PDFBox, 4.0.0
>
> Attachments: EmptyActualText_poppler.txt,
> EmptyActualText_reduced_poppler.txt, Main.java,
> PDFBOX-5868-7FHMU2HNOUUPENUPKZDGD2V65YEVABRS-EmptyActualText.pdf,
> PDFBOX-5868-SI5K4X4Z55SQAUPLAUP6QRRWT3UD3LAA-EmptyActualText.pdf,
> PDFBOX-5868-SI5K4X4Z55SQAUPLAUP6QRRWT3UD3LAA-EmptyActualText_reduced.pdf,
> Tilman's_solution_out.txt, adobe_out.txt,
> content_diffs_with_exceptions-ActualText.xlsx,
> image-2024-08-19-10-38-13-472.png, image-2024-08-30-17-55-41-423.png,
> multilingual_test.pdf, okular_out.txt, page.pdf, pdfbox_out.txt,
> poppler_out.txt, screenshot-1.png, screenshot-2.png,
> suppressDuplicateOverlapping_out.txt
>
>
> I downloaded the latest executable jar of pdfbox (3.0.3) for testing and used
> the export:text command line tool to obtain the results
> * the multilingual_test.pdf is the original pdf i made to test multilingual
> text extraction.
> * the pdfbox_out.txt is the text file produced by pdfbox
> * the adobe_out.txt is the text file created by adobe reader's save as text
> feature
>
> Observation:
> as you can see in the attachment the text file obtained by pdfbox shows weird
> unicodes for tamil and bengali (for hindi the charecters are extracted but
> not overlapped; japanese seems fine to me). in contrast the text file file
> obtained from adobe reader's save as text feature seems fine and copy pasting
> the text from my document viewer(evince) also works.
> Questions:
> # why are the outputs from pdfbox and adobe different?
> # what can i do to extract the text from a multilingual pdf correctly?
> # Is there a way to apply pattern matching to text in pdf file and declare
> matches without extracting the text first? (say if the problem is with fonts
> and glyphs)
> —
> My Usecase fyi:
> i am trying to extract text from files and run pattern matching. I am using
> apache tika for parsing documents. I noticed problem with extracted PDF text
> (other filetypes parse fine). used executable pdfbox jar to conclude that the
> _problem is in pdfbox and not in tika._ tested with adobe reader's extract
> text to confirm the problem is not with the pdf. i want to extract these
> multilingual text to run pattern matching on them alone and do not need to
> display the content but only if the pattern is present or not (say if the
> problem is with fonts and glyphs)
>
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscr...@pdfbox.apache.org
For additional commands, e-mail: dev-h...@pdfbox.apache.org
