Dear Merlijn, Thank you for your efforts in providing the files "Samle-out.pdf" and "Samle-out.pdf2". I have checked these files. The file Samle-out2.pdf highly compressed with good quality. Kindly share the soiurce code or the tools that you have used to generate the file "Samle-out2.pdf". Kindly let us know that whether i can use your source code or tools on Windows 10 PC.
Thanks and Regards, Subramanyam On Wednesday, April 14, 2021 at 10:21:14 PM UTC+5:30 Merlijn Wajer wrote: > Hi, > > On 14/04/2021 15:26, Sharp Subbu wrote: > > Dear Merlijn, > > > > Thank you very much for your reply. > > We are doing feasibility study on using Tesseract OCR featurs in our > > project on Windows 10 English 32/64-bit OS. > > As part of this study, i am trying to find that is it possible to > compress > > / reduce the size of the pdf file created by Tesseract OCR (CommandLine: > > > > Tesseract input.tif outputFile pdf). > > To find answer for this question, I have checked tesseract forums, and > > Tesseract APIs. I did not find any related information. Hence, I have > > posted the same question in Tesseract Google forums. > > Regarding this, i received nice reply from you. Thank you very much for > > that. > > Firstly, clarify that is Tesseract OCR API supports reducing / > compressing > > the OCRed pdf file. Is this support present or not in Tesseract OCR > sourc > > code. > > > > Kindly fin dthe attached sample pdf file "Sample.pdf" for your > reference. > > Kindly compress it and send the compressed pdf file. > > Please see attached 'sample-out.pdf' (compression ratio ~3.5), and > 'sample-out2.pdf' (compression ratio ~7.5). > > I also generated one other PDF which illustrates how the compression > works, rather than being actually compressed, but that file is not > attached due to file size reasons (~460KB). You can that file (plus the > other two files) here: https://archive.org/~merlijn/tmp/mrc-pdf/ > > Please see the commands I used: > > 1. Extract image from PDF (not necessary if you start with an image, > even better to start with the image and not have Tesseract generate the > PDF) > > > $ pdfimages -all /tmp/Sample.pdf /tmp/sample > > 2. OCR: > > > $ tesseract /tmp/sample-000.jpg - hocr > /tmp/sample-000.hocr.html > > 3. Create PDF (dpi taken from JPEG image): > > > $ PATH=$PATH:/home/merlijn/archive/pdf/bin recode_pdf --from-imagestack > /tmp/sample-000.jpg --hocr-file /tmp/sample-000.hocr.html -o > /tmp/sample-out.pdf -m 2 -v --dpi 200 > > [...] > > Processed 1 pages at 1.80 seconds/page > > Compression ratio: 3.677732 > > 4. Create more compressed PDF (ditto for dpi): > > > $ PATH=$PATH:/home/merlijn/archive/pdf/bin recode_pdf --from-imagestack > /tmp/sample-000.jpg --hocr-file /tmp/sample-000.hocr.html -o > /tmp/sample-out2.pdf -m 2 --dpi 200 --bg-downsample 3 -v > > [...] > > Processed 1 pages at 1.59 seconds/page > > Compression ratio: 7.581879 > > Size comparison: > > > $ ls -lsh /tmp/sample-out*.pdf /tmp/Sample.pdf > > 40K -rw-r--r-- 1 merlijn merlijn 38K Apr 14 18:08 /tmp/sample-out2.pdf > > 80K -rw-r--r-- 1 merlijn merlijn 79K Apr 14 18:11 /tmp/sample-out.pdf > > 296K -rw-r--r-- 1 merlijn merlijn 293K Apr 14 17:56 /tmp/Sample.pdf > > Note that: > > 1. Compression is not lossless, but text should nevertheless be quite > sharp. > 2. I run Tesseract on the image in the PDF, so for your purposes you > might want to instead generate hOCR from the image and let 'recode_pdf' > make the PDF for you (it's pretty much the same code that Tesseract uses). > 3. The mask in the PDF is encoded with 'ccitt' and not 'jbig2', which > would give you slightly better compression still. (This is a bug in > mupdf which will be fixed in the next mupdf release, I have a patched > version somewhere, but not at hand) > 4. I have only tested this on Linux. > 5. The above run uses Kakadu for JPEG2000 compression, but you could > also use Grok [0] or OpenJPEG [1] (OpenJPEG already works as per my > previous email). > > Finally, for some reason it looks like the PDFs created with the recode > tool actually look better than the sample you sent me -- I think that is > because yours suffers from JPEG artifacts which gets mostly cancelled > out by the mask technique that MRC employs. > > If this looks like something you might want to use, we could talk > off-list about how to make it works on Windows, to not bother the list > with details not relevant to Tesseract. > > Cheers, > Merlijn > > [0] https://github.com/GrokImageCompression/grok/ > [1] https://www.openjpeg.org > > > Thank you very much for your nice help. > > Subramanyam > > > > > > On Wednesday, April 14, 2021 at 6:27:43 PM UTC+5:30 Merlijn Wajer wrote: > > > >> Hi, > >> > >> On 14/04/2021 13:52, Sharp Subbu wrote: > >>> Dear friends, > >>> > >>> Kindly guide/help us to find solution for the below point: > >>> ============================= > >>> How to reduce the size of a OCRed pdf file using Tesseract OCR APIs. > >>> =============================== > >> > >> Not sure exactly what use case you have in mind (OS, etc), but I have a > >> suggestion, as I dealt with this in the recent past. > >> > >> I developed something similar to the foxit/luratech "PDF compression", > >> in Python and it is entirely open source. It uses the Tesseract hOCR > >> result files. The can lead to 3-15x compression ratios (sometimes more, > >> depending on the image formats that you use). > >> > >> It converts images to JPEG2000 for best compression (but slower loading > >> times) and also attempts to create a "foreground", "background" and > >> "mask" image (Mixed Raster Content [0]), which can significantly > improve > >> compression. It inserts a text layer just like Tesseract does (the code > >> is a port of Tesseract's C++). > >> > >> Here is some info [1], and here is the source code [2]. > >> > >> There is a "openjpeg-wip" branch that can use OpenJPEG instead of > Kakadu > >> for image compression. > >> > >> Example usage to create a PDF from a set of images: > >> > >> recode_pdf --from-imagestack 'images/*.jp2' --hocr-file > >> combined_tesseract_results.html -o out.pdf -v --use-openjpeg -m 2 > >> > >> There is also the --from-pdf option instead of --from-imagestack, but > >> that has only seen light testing. > >> > >> You can combine the hOCR result files using hocr-combine-stream [3] > >> > >> If this suits your use case, I'd be happy to help/assist here or off > >> list. There aren't many users of the software yet (the same offer > >> extends for others reading this list). If you have an example PDF that > >> you can send me, I'd be happy to try to send you a compressed PDF back. > >> > >> Cheers, > >> Merlijn > >> > >> > >> [0] https://en.wikipedia.org/wiki/Mixed_raster_content > >> [1] https://archive.org/~merlijn/projects/archive-pdf-tools/index.html > >> [2] https://git.archive.org/merlijn/archive-pdf-tools > >> [3] > >> > >> > https://git.archive.org/merlijn/archive-hocr-tools/-/blob/master/bin/hocr-combine-stream > > >> > > > > -- You received this message because you are subscribed to the Google Groups "tesseract-ocr" group. To unsubscribe from this group and stop receiving emails from it, send an email to tesseract-ocr+unsubscr...@googlegroups.com. To view this discussion on the web visit https://groups.google.com/d/msgid/tesseract-ocr/3fdbb11c-8f88-49e7-b824-6a793bc29cc5n%40googlegroups.com.
