PDF support

[MichaelF25]

Hi,

are there any plans to add application/pdf as supported Content-type within the multipart/related request?
Maybe it's enough to add ghostscript to the tesseract docker image (if it's not already present) and adjust the Content-type check in the http handler.

[tleyden]

No plans but it sounds like a useful feature

Does ghostscript just render pdf -> image?

[MichaelF25]

To be honest, I am not sure. I've just evaluated tesseract via command line for my use case (OCR for my PDFs) and it worked out of the box if ghostscript was also installed on the system. This leads me to my initial assumption above.

[parkerhancock]

I'll second this as a feature request. Ghostscript compiles pdf -> postscript, but there are lots of tools that can convert the .ps file to .png, .tiff, etc.

Honestly, if you're using PDF's, chances are the PDF is just a wrapper for .TIFF images that can be directly extracted without any resampling/processing. Python's PyPDF2 package can retrieve these:

https://stackoverflow.com/questions/2693820/extract-images-from-pdf-without-resampling-in-python

Also, by far the most common PDF document format is embedded Group 4 Tiffs. These can be picked out of pdf's just by finding the right byte offsets and adding a correct .TIFF header and tags. This process is super-fast, and PyPDF2 can do it quite quickly.

But, Tesseract is very picky about taking input from .tiff files, and frequently throws errors if the tags aren't right in the image. I've struggled to extract tiffs that Tesseract will handle from PDF's. Ghostscript or ImageMagick are good ways to make .tiffs or other images that make Tesseract happy, but both are pretty substantial dependencies to add to an application. And slow the process way down.

Unfortunately, Go is complete greek to me, so I don't know how to do something similar in Go. Alternatively, are there any plans or ideas on how to integrate python preprocessor modules? This seems like a good candidate. I could (and already am) taking PDFs of government documents, stripping out tiffs in my applications. OCR is just the one missing step (and why I landed here). I need to add OCR capabilities to a web application, and need a cloud platform.

[parkerhancock]

Update: Much faster image extraction can be done with poppler using the pdfimages command. And Tesseract doesn't complain about the output.

[sferoze]

@Afterburned6387

I would like to also pass PDF document links to openOCR. Like so:

curl -X POST -H "Content-Type: application/json" -d '{"pdf_url":"http://bit.ly/ocrpdf","engine":"tesseract"}' http://IP_ADDRESS_OF_DOCKER_HOST:HTTP_PORT/ocr

Just instead of img_url in the body it is pdf_url

Anyone have a rough outline of how I would implement this addition?

[tleyden]

@sferoze the current infrastructure in place for this is to create a preprocessor like the Stroke Width Transform and add it to the pipeline. This would require some coding.

[TukaramKanade]

Hi All,
Any update on the PDF support, is it included ? If yes , could you please provide the details on it.

[tleyden]

@TukaramKanade not that I'm aware of

[xf0e]

Since tesseract 4 has no support for reading pdf files, and we need to handle them i managed to create a new preprocessor convert-pdf for this purpose.

https://github.com/xf0e/open-ocr/blob/master/convert-pdf.go

It will convert an incoming PDF file with ImageMagick an pass it to tesseract_engine.

I hope it helps.

[tleyden]

@xf0e that looks useful! Want to send a PR? There's no CI in place right now, but I will hand-verify it.

[xf0e]

@tleyden Done! #108