[
https://issues.apache.org/jira/browse/TIKA-3270?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17264228#comment-17264228
]
Luís Filipe Nassif commented on TIKA-3270:
------------------------------------------
Is checking for missing ToUnicode mapping easy? I have never seen this part of
the code. And if we also could detect garbage text with lang stats as suggested
by [~tilman] , I think this would be a great improvement to OCR of PDFs! Thanks
for opening this, [~tallison]!
> Render non-text in PDFs for OCR
> -------------------------------
>
> Key: TIKA-3270
> URL: https://issues.apache.org/jira/browse/TIKA-3270
> Project: Tika
> Issue Type: Improvement
> Reporter: Tim Allison
> Priority: Major
> Attachments: tiger.pdf
>
>
> When we render a PDF page for OCR, we are relying on PDFBox to render all of
> the contents of the page, including text that may be available via regular
> extraction methods.
> The result of this is that if a user selects ocr_and_text, there can be
> duplicate text -- text as stored in PDFs and the text generated via OCR. In
> the xhtml output, we do mark a separate "div" for OCR so that users can
> distinguish, but still, it might be useful not to have to run OCR on text
> that was reliably extracted.
> One solution to this was proposed by [~lfcnassif] on TIKA-3258, with a
> technical/implementation recommendation by [~tilman] to subclass PDFRenderer
> and PageDrawer to render only the image components of a page.
> This would be a new, non-breaking feature. This is not a blocker on 2.0.0.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
