Hi Cameron, I am guessing your cluster software already assigns an instance of tesseract to each core, Olena might be useful on a single machine with multiple cores for a single image, but it sounds like you are way beyond that scenario. I don’t know if it would be worth using Olena to remove illustrations since it has some overhead itself in figuring out the segmentation. Not a lot, but I wonder if it would eat any savings you might see. Sorry, I don’t know anything about Tesseract-4.0, good luck with your processing.
art From: [email protected] [mailto:[email protected]] On Behalf Of Cameron Fen Sent: Friday, January 27, 2017 1:34 PM To: tesseract-ocr <[email protected]> Subject: Re: [tesseract-ocr] improving performance (and speed) of tesseract Hi Art, Thanks! I am already using multiprocessing on a cluster as I have at least 100k images to OCR. Correct me if I'm wrong, but if I already have a full load each working on a different image, using Olena to carve the image into paragraphs will not improve speed. On another note, we don't need illustrations, but if we can remove all the extraneous lines and pictures with Olena will that help accuracy? I will take a look at the wiki. Also does anyone know if it is worth it to try and use Tesseract 4.0 given the headache of installing it? Thanks, Cameron On Friday, January 27, 2017 at 12:53:13 PM UTC-5, Art Rhyno wrote: I think pytesseract is a wrapper to the command-line, so you would probably see the same results by going directly. The python-tesseract [1] project used swig to do a deeper level of integration, though I tried the same approach a few years ago and didn’t really notice much difference in throughput. It’s possible you could use a segmentation tool like Olena [2] to carve up the image into individual paragraphs and fire off multiple instances of tesseract (see discussion here [3]). Olena might also give you a way to extract illustrations and such. As for accuracy, I tried a Gaussian blur via opencv on the image and got slightly better results, though comparing OCR can be dicey past a certain level (the blur gave a few more words but also messed up a number). There are lots of tips in the mailing list on improving accuracy by image manipulation, but I’d start with the wiki page on this topic [4]. art --- 1. https://code.google.com/archive/p/python-tesseract 2. http://olena.lrde.epita.fr 3. http://stackoverflow.com/questions/4962978/is-tesseract-3-00-multi-threaded 4. https://github.com/tesseract-ocr/tesseract/wiki/ImproveQuality . -- 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 [email protected]<mailto:[email protected]>. To post to this group, send email to [email protected]<mailto:[email protected]>. Visit this group at https://groups.google.com/group/tesseract-ocr. To view this discussion on the web visit https://groups.google.com/d/msgid/tesseract-ocr/a8aa3985-a8ad-4032-894b-7d9782728aba%40googlegroups.com<https://groups.google.com/d/msgid/tesseract-ocr/a8aa3985-a8ad-4032-894b-7d9782728aba%40googlegroups.com?utm_medium=email&utm_source=footer>. For more options, visit https://groups.google.com/d/optout. -- 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 [email protected]. To post to this group, send email to [email protected]. Visit this group at https://groups.google.com/group/tesseract-ocr. To view this discussion on the web visit https://groups.google.com/d/msgid/tesseract-ocr/DM3PR11MB103669CB133D3EE449446373DC760%40DM3PR11MB1036.namprd11.prod.outlook.com. For more options, visit https://groups.google.com/d/optout.
