You are viewing a plain text version of this content. The canonical link for it is here.
Posted to user@tika.apache.org by Stefan Alder <tw...@gmail.com> on 2015/05/13 21:03:49 UTC

Embedded images in PDF - detect, extract and/or OCR

Ultimately I'm trying to (1) determine whether images, particularly, full
page images, are embedded in a pdf, and (2) extract the images and/or (3)
OCR the text.

Does tika-app support this?  When I run java -jar tika-app-1.8.jar
test.pdf, I get all of the meta data, and see <page></page> tags but no
images.

Running with -z doesn't output any images.

RE: Embedded images in PDF - detect, extract and/or OCR

Posted by "Allison, Timothy B." <ta...@mitre.org>.
Hi Stefan,


1)      Right, out of the box, tika-app does not provide information about whether an embedded/inline image exists.  It will handle “attached” images as all other parsers do out of the box, but not embedded/inline images.

2)      Disabled for inline images, but not for regular attachments.

3)      At this point, no.  One hack is to unzip the app jar and just change the values in the properties file and rezip the jar.  On the horizon, I’d like to make a common interface for parser configuration so that you can set parser config parameters via the regular tika config file, and then you’d be able to specify that at the commandline.


If you do change the properties file, you’ll probably also want to change extractUniqueInlineImagesOnly
to “false”.

Cheers,

          Tim

From: Stefan Alder [mailto:twigbranch@gmail.com]
Sent: Wednesday, May 13, 2015 3:30 PM
To: user@tika.apache.org
Subject: Re: Embedded images in PDF - detect, extract and/or OCR

To clarify,
(1) tika-app, as compiled, does not provide any indication that an image exists within a pdf? (my main interest are entire page images for PDFs that were scanned). Again, my first interest is detecting whether embedded images exist.
(2) the -z option is effectively disabled for PDFs?
(3) is there a way to enable detection and/or extraction from the command line, as opposed to editing the source?



On Wed, May 13, 2015 at 12:18 PM, Allison, Timothy B. <ta...@mitre.org>> wrote:
By default, Tika is configured not to extract embedded images from PDFs because in some edge cases, there can be thousands of images in some small PDF files (see https://issues.apache.org/jira/browse/TIKA-1294).  Our choice to have the default be “don’t extract” was based on the concern that if we made the change, devops folks in large document processing pipelines might be surprised by memory consumption and far slower parsing.

To configure Tika to extract embedded images, you can configure a PDFParserConfig (setExtractInlineImages(true)) and attach that to a ParseContext before the parse, or (if you are just using tika-app) you can set that value manually in in the app jar in o.a.t.parser.pdf.PDFParser.properties.

I’m haven’t tested whether our OCR parser will process those embedded images, but it should.

Let me know if this helps.

From: Stefan Alder [mailto:twigbranch@gmail.com<ma...@gmail.com>]
Sent: Wednesday, May 13, 2015 3:04 PM
To: user@tika.apache.org<ma...@tika.apache.org>
Subject: Embedded images in PDF - detect, extract and/or OCR

Ultimately I'm trying to (1) determine whether images, particularly, full page images, are embedded in a pdf, and (2) extract the images and/or (3) OCR the text.

Does tika-app support this?  When I run java -jar tika-app-1.8.jar test.pdf, I get all of the meta data, and see <page></page> tags but no images.

Running with -z doesn't output any images.




Re: Embedded images in PDF - detect, extract and/or OCR

Posted by Stefan Alder <tw...@gmail.com>.
To clarify,
(1) tika-app, as compiled, does not provide any indication that an image
exists within a pdf? (my main interest are entire page images for PDFs that
were scanned). Again, my first interest is detecting whether embedded
images exist.
(2) the -z option is effectively disabled for PDFs?
(3) is there a way to enable detection and/or extraction from the command
line, as opposed to editing the source?



On Wed, May 13, 2015 at 12:18 PM, Allison, Timothy B. <ta...@mitre.org>
wrote:

>  By default, Tika is configured not to extract embedded images from PDFs
> because in some edge cases, there can be thousands of images in some small
> PDF files (see https://issues.apache.org/jira/browse/TIKA-1294).  Our
> choice to have the default be “don’t extract” was based on the concern that
> if we made the change, devops folks in large document processing pipelines
> might be surprised by memory consumption and far slower parsing.
>
>
>
> To configure Tika to extract embedded images, you can configure a
> PDFParserConfig (setExtractInlineImages(true)) and attach that to a
> ParseContext before the parse, or (if you are just using tika-app) you can
> set that value manually in in the app jar in
> o.a.t.parser.pdf.PDFParser.properties.
>
>
>
> I’m haven’t tested whether our OCR parser will process those embedded
> images, but it should.
>
>
>
> Let me know if this helps.
>
>
>
> *From:* Stefan Alder [mailto:twigbranch@gmail.com]
> *Sent:* Wednesday, May 13, 2015 3:04 PM
> *To:* user@tika.apache.org
> *Subject:* Embedded images in PDF - detect, extract and/or OCR
>
>
>
> Ultimately I'm trying to (1) determine whether images, particularly, full
> page images, are embedded in a pdf, and (2) extract the images and/or (3)
> OCR the text.
>
>
>
> Does tika-app support this?  When I run java -jar tika-app-1.8.jar
> test.pdf, I get all of the meta data, and see <page></page> tags but no
> images.
>
>
>
> Running with -z doesn't output any images.
>
>
>
>
>

RE: Embedded images in PDF - detect, extract and/or OCR

Posted by "Allison, Timothy B." <ta...@mitre.org>.
By default, Tika is configured not to extract embedded images from PDFs because in some edge cases, there can be thousands of images in some small PDF files (see https://issues.apache.org/jira/browse/TIKA-1294).  Our choice to have the default be “don’t extract” was based on the concern that if we made the change, devops folks in large document processing pipelines might be surprised by memory consumption and far slower parsing.

To configure Tika to extract embedded images, you can configure a PDFParserConfig (setExtractInlineImages(true)) and attach that to a ParseContext before the parse, or (if you are just using tika-app) you can set that value manually in in the app jar in o.a.t.parser.pdf.PDFParser.properties.

I’m haven’t tested whether our OCR parser will process those embedded images, but it should.

Let me know if this helps.

From: Stefan Alder [mailto:twigbranch@gmail.com]
Sent: Wednesday, May 13, 2015 3:04 PM
To: user@tika.apache.org
Subject: Embedded images in PDF - detect, extract and/or OCR

Ultimately I'm trying to (1) determine whether images, particularly, full page images, are embedded in a pdf, and (2) extract the images and/or (3) OCR the text.

Does tika-app support this?  When I run java -jar tika-app-1.8.jar test.pdf, I get all of the meta data, and see <page></page> tags but no images.

Running with -z doesn't output any images.