Entire documents are nothing but images

[Giovanni290542994dvw]

Hi,

I was trying to extract the document attached to this question, when I try to extract it with Adobe APIs I just get a collection of images, one for each page.

I think this has to do with document parsing (probably parsed as a svg) and I don't know how to solve it!

Can sombody help me?

Thanks,

Giovanni

[Raymond Camden]

To be cleare, are you using the Extract API? It doesn't return just an image, but JSON, and optionally included images in the PDF (as well as other stuff).

[Giovanni290542994dvw]

Yes, I am using the Extract API.

I only get the image data from each page, can you send me the JSON you are talking about?

thanks in advance,
Giovanni

[Raymond Camden]

What you are describing is impossible. 🙂 Our Extract API returns a zip file. The zip file _always_ contains structuredData.json. It *optionally* includes images and tables. 

[Giovanni290542994dvw]

Is there a way to extract the JSON representation for each vector Figure element?

[Giovanni290542994dvw]

For reference, this is the output I get:

 

[Raymond Camden]

I don't understand your output. See my earlier comment. The API returns a zip, not just an image. Maybe show the code you're using?

[Giovanni290542994dvw]

I indeed get a zip file, but for some reason each page in this particular document is extracted as if it was a Figure and not as structured data.
The code I am using works well for any documents (and is identical to the examples provided in the documentation) but in this case is unable to detect the structured data contained in pages.

To better understand my output simply upload the document attached in the Extract API.

[Aftia_Jeff]

@Giovanni290542994dvw   I get the same result as you do when I run your file through the SDK.   There is no text in the resulting JSON.   We've had similar issues with some files where numerous lines and boxes are recognized as an image instead of text.  However, your PDF seems to be mostly text-based.

[Giovanni290542994dvw]

Hi @Aftia_Jeff, did you guys find a workaround for this issue?

[Aftia_Jeff]

No, the issue we have run into is a product bug (DCSV-53202) which is because the ML model wasn't trained on forms (which contain a lot of lines and boxes).   However, your example document doesn't look like a form.

[Raymond Camden]

Ok, to be clear, you are getting the JSON back - I tested via the SDK and I see the JSON. And it has structured data. But it does not seem to read the text. If we look at the properties of the PDF, the producer, iLovePDF, is not a great one. So yes you have a readable PDF, but the underlying way it was built was "less than optimal" (imo). It's basically really a set of images so Extract is 'properly' reading it. Best I can suggest is creating your PDF in a better tool (obviously we would recommend Acrobat. 😉

[Giovanni290542994dvw]

@Raymond Camden Thanks for the response.

In such cases is it possible to "convert" the PDF in a more readable format?

[Raymond Camden]

Per my coworker, Joel Geraci, who is a PDF Jedi, he said he was able to use Acrobat to export to PostScript, convert it back to PDF, and it corrected it. If that's an option for you, you can consider that.

[Giovanni290542994dvw]

that is interesting, thanks!