FREE OCR - ALL COLLECTIONS -TYPED AND HANDWRITTEN TEXTS

Thank you Rodrigo, our Library of Congress colleague who kindly agreed to share his OCR method here on History Hub for volunteers to use if they like. This process will be helpful for getting text for printed documents, rather than handwritten ones. It can really save time when you're dealing with newsprint, though as Rodrigo says and demonstrates in his examples, there are times when OCR doesn't work so well--always read through the automatically generated text to check that it accurately represents the original (including any punctuation, accents or spelling, regardless of whether this is "correct" by modern standards). You can also get images of each document by clicking on the "view on loc.gov" button in the transcription interface.

Please preserve the original line breaks as you would if you were transcribing. Make sure that words broken over two lines such as kit-

ten

are transcribed as a whole word on the first line, so "kitten", in this example.

I have started using this OCR method and it is awesome. Thanks so much for sharing.

The OCR reader I have been using with good success on the iPad is called HRReader (free in Apple App Store). Need to search for “Tahira Ghani ocr” it doesn’t come up under HRReader. It claims to read handwriting too but I haven’t had much luck with that. But it does work well with printed text. As was stated above, you still need to proofread mostly for misspelled words before passing the document on for review.

I have come across many documents that were waiting for review that were not proofread and had to fix for minor misspellings (on instead of “of”, etc.)

BTW, when OCR’ing a multi column page, I do it one column at a time.

Pomodoro software makes this process easier.

https://pomodoro.semlab.io/

Only an upload is mandatary.

Can someone please explain this procedure?  Where do you get All Collections?  Then you do something in Google Drive?  That sounds cloud base which is not viable for someone stuck with DSL.  I need to process about 15,000 pdfs of 5 to 20 pages.  They were OCR'ed at one time but the company that did it messed up the scanning resolution (every other page is off by a factor of 4) and the cropping.  I would say 90% of the text is jibberish.  Will I need to extract every pdf page as a jpg, do a new OCR, and then rebundle the individual pdfs? 

Hi Michael,

The original message gave instructions for volunteers working on the By the People project, which is the Library of Congress' crowdsourcing transcription program so they may not apply to your situation. "All collections" refers to Library of Congress materials. Our platform doesn't display PDFs and we're usually dealing with files that only include 1 page, which is why this method is pretty specific to the file types/sizes encountered on the Library of Congress sites. There may be volunteers on this message board who could provide advice on your situation, but you might also consider reaching out to the organization or department (if they are Library of Congress materials) where you sourced the PDFs to find out what they advise.

Best,

Abby

By the People Community Manager

Thank you for your reply.  Yes, I can see now that the post is in reference to a specific project.  My read was that "All Collections" might be a software app to perform OCR.  I've been struggling with various programs, all results are marginal to horrid.  As for the source of the pdfs, they are from a City archive.  I was able to get the pdfs, I doubt they have the original jpgs.  Whoever did the scanning had every other page set at a different resolution.  As you view the pdf, it goes from too big to too small.  They also made a mess of file naming, listing month and day prior to year.  The City has a web portal that allows for searching, but the user is limited to single word searches.  That, coupled with poor OCR, means you seldom get a meaningful search.

I thought it would be an interesting project to fix the file names (done), and the page formatting (done).  I then wanted to obtain a new OCR scan of the documents and create a searchable index that would allow searching for phrases, not just single words.  The original documents are in very large ledger style books and I'm afraid they will all need to be reshot to make this happen.  The City archive has been very helpful in my research and I took this on as a way to pay back their help.

Hi Michael,

plese try Pomodoro https://pomodoro.semlab.io/

Here you can check how it works https://pomodoro.semlab.io/

In this site you can discard what you don't need and then create a Text file.

You alco can upload all the 15 thousand files to Google drive than open each one as Google document.

Thank you so much!  I will look into this.  I may need to be thinking about this being a crowd source project since there is so much to process.  Given the image quality and format issues of this collection, transcription will likely be the only way to get a reliable search capability.  Thanks again.