[sdiy] Can anyone OCR the AN23.PDF File Here?

Thu Jul 6 21:41:34 CEST 2017

Bernie,

I'd be happy to have a go using open source software such as tesseract. I
feel you cannot tackle this problem without further teaching the classifier
for the nuances of this text and constraining of what the possibilities
are. I'd like you to rescan a representative item into tiffs at 300 or
600dpi. You stated that the tiffs would look the same but that is not true.
The tiff exports out of AN23.pdf do not have the same quality of the
original scanned image (likely tiffs), and this does not help.

Is the end goal to replace the original image text with just text using a
similar font? For me the most useful would be to have OCR'ed most of it so
it's searchable. But again, you have not set what you find acceptable or
what your goal is. It's ok if you don't know. You either propose this as a
challenge that will never be possible to accomplish (some teachers never
want to give students top score) or you compromise and propose a solution
that enhances your original pdfs with value worth money for most people. I
find most people would be happy keeping the original scanned text and
simply having it OCR'ed to the best possible for their cursory searches.
But I may see things different than most people :) In the academic lingo
you should provide some ground-truth examples of what you imagine is a
perfect conversion of the AN23.pdf, acceptable or not worth the time.

You can never rely on ONE volunteer, but you can certainly get many excited
so over time as a group something is accomplished.

Cheers
b

On Thu, Jul 6, 2017 at 3:11 PM Bernard Arthur Hutchins Jr <bah13 at cornell.edu>
wrote:

> Thanks Rob -
>
>
> Really makes my point, and I guess I should not rely on volunteers!  I
> don't blame you one bit - just does not work.
>
>
> I expect no one else want to try either.  If anyone does, don't look at
> the crib below until after you try.   Errors located and circled in red.
>
>
> http://electronotes.netfirms.com/AN23Rob.PDF
>
>
> Please all, let's agree that the OCR issue is bogus as applied here.
>
>
> Bernie
>
>
> ------------------------------
> *From:* Rob Kam <robkam at ymail.com>
> *Sent:* Thursday, July 6, 2017 1:51 PM
>
> *To:* Bernard Arthur Hutchins Jr
> *Cc:* synth-diy at synth-diy.org
> *Subject:* Re: [sdiy] Can anyone OCR the AN23.PDF File Here?
>
> Thanks for the challenge Bernie but no thanks. I don't have the patience
> to correct the OCR.
>
> Rob
>
> ------------------------------
> *From:* Bernard Arthur Hutchins Jr <bah13 at cornell.edu>
> *To:* Rob Kam <robkam at ymail.com>
> *Cc:* "synth-diy at synth-diy.org" <synth-diy at synth-diy.org>
> *Sent:* Thursday, 6 July 2017, 18:30
>
> *Subject:* Re: [sdiy] Can anyone OCR the AN23.PDF File Here?
>
> Thanks Rob -
>
> True - the equations are now usable, but slightly more blurred than my
> original PDF. Likewise, the figures are now OK but of slightly lower
> quality, which does NOT matter much for hand drawings.
>
> I did note a lot of OCR misreads in the text.  A careful proofing of the
> text took me 18 minutes and there are 25 errors, some not at all obscure,
> and about 13 of which I had to look at the original to see what they were
> supposed to be.  (One was hard to detect since it substituted an Rf for an
> Ri, a disaster).  A full proofread/correction would take at least
> 30 minutes (188 eight-hour days for 6000 pages).    And I wrote this!
> Almost certainly a volunteer would have more trouble and miss errors.
>
> In the spirit of no good deed going unpunished, Rob, let me put you on the
> spot. Take your scan, find and fix the 25 errors.  Let us know how
> easy/hard this was and the time it took, and show your results.
>
> I will post the "solution" to the "find the errors" this evening if I get
> the chance.
>
> Since there is no improvement in the figures/equations, and the text is a
> serious downgrade, tell me again (anyone) why an OCR/ebook is a good idea
> here.
>
> Bernie
>
>
> ------------------------------
> *From:* Rob Kam <robkam at ymail.com>
> *Sent:* Thursday, July 6, 2017 7:24 AM
> *To:* Bernard Arthur Hutchins Jr
> *Cc:* synth-diy at synth-diy.org
> *Subject:* RE: [sdiy] Can anyone OCR the AN23.PDF File Here?
>
> There’s a second attempt at http://www.sdiy.info/AN23b.rtf converting the
> equations to images instead, (and still manually tweaking the OCR). It took
> six minutes to do from the scan/PDF and the text still needs comparing and
> correcting against the original.
>
> There are already experts at this sort of project, at Archive.org who have
> been doing this for a number of years
> https://archive.org/details/texts&tab=about
> Free Books : Download & Streaming : eBooks and Texts ...
> <https://archive.org/details/texts&tab=about>
> archive.org
> The Internet Archive offers over 12,000,000 freely downloadable books and
> texts. There is also a collection of 550,000 modern eBooks that may be
> borrowed by anyone ...
>
> Free Books : Download & Streaming : eBooks and Texts ...
> <https://archive.org/details/texts&tab=about>
> archive.org
> The Internet Archive offers over 12,000,000 freely downloadable books and
> texts. There is also a collection of 550,000 modern eBooks that may be
> borrowed by anyone ...
>
>
>
> To put my two cents in, the synth DIY community should see whether they
> are able to raise the funds to compensate (against unsold hardcopy, ebooks
> etc.) for releasing Electronotes under a non-commercial Creative Commons
> licence https://creativecommons.org/licenses/by-nc/2.0/uk/
>
> Rob
>
> *From:* Bernard Arthur Hutchins Jr [mailto:bah13 at cornell.edu]
> *Sent:* 06 July 2017 01:42
> *To:* Rob Kam <robkam at ymail.com>; mskala at ansuz.sooke.bc.ca
> *Cc:* synth-diy at synth-diy.org
> *Subject:* Re: [sdiy] Can anyone OCR the AN23.PDF File Here?
>
>
> Tkanks Rob -
>
> But a manual identifications and 5 minutes/page is no good for the small
> improvement. Still months of 8-hour days to do 6000 pages.  My PDF is still
> much better already.  The equations are still unusable.  It makes the same
> text errors, apparently.    Why not just say it just can't do this?
> Wasn't intended to.
>
> Thanks for trying - useful data point!
>
> Bernie
> ------------------------------
> *From:* Rob Kam <robkam at ymail.com>
> *Sent:* Wednesday, July 5, 2017 6:47 PM
> *To:* Bernard Arthur Hutchins Jr; mskala at ansuz.sooke.bc.ca
> *Cc:* synth-diy at synth-diy.org
> *Subject:* RE: [sdiy] Can anyone OCR the AN23.PDF File Here?
>
> Hi Bernie,
>
> At http://www.sdiy.info/AN23.rtf this took 10 minutes to OCR with ABBYY
> FineReader 12
> <https://www.google.co.uk/url?sa=t&rct=j&q=&esrc=s&source=web&cd=1&cad=rja&uact=8&ved=0ahUKEwiZhc6ZmPPUAhVG6RQKHRHpA1UQFggoMAA&url=http%3A%2F%2Fwww.abbyy.com%2Fen-gb%2Fsupport%2Ffinereader-12%2F&usg=AFQjCNHLOjsz219pjjTDqDytG2Cpm9N90w>,
> first manually identifying areas of text vs. images. Obviously it still
> needs further corrections.
>
> Rob
>
>
> _______________________________________________
> Synth-diy mailing list
> Synth-diy at synth-diy.org
> http://synth-diy.org/mailman/listinfo/synth-diy
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://synth-diy.org/pipermail/synth-diy/attachments/20170706/675585cc/attachment.htm>