How can I legally convert a “dead-tree” library (1050 books) to eBooks? • Hacker News Board

Re: How can I legally convert a “dead-tree” library (1050 books) to eBooks?

#11

Post by alok-g » Sun, May 31, 2015, 1:57 AM UTC

walterbell wrote: post #6

vitovito wrote: post #3

They're your books, so you can scan them if you want. But you can't sell, donate or give away the physical book once you've scanned it; the scan is a copy/backup of the original for personal use, and if you get rid of the original, you have to delete the copies, too. You can only destroy the original. There are bulk, destructive book scanners which will give you PDFs of your books. You don't get the books back, becau…

Is there any way to buy a kit for a nondestructive scanner? 1dollarscan is rumored to be watermarking/encoding the customer name in their scans. If you highly magnify the page image, there are a large number of small artifacts, which may explain the large file sizes, e.g. 150MB for a 400 page book. Their service ($1 per 100 pages) is cost-effective, given that Staples will charge $2 to remove the spine from a book, w…

>> 1dollarscan is rumored to be watermarking/encoding the customer name in their scans.

I rather appreciate that as a means against piracy of the scanned books.

Re: How can I legally convert a “dead-tree” library (1050 books) to eBooks?

#12

Post by wodenokoto » Sun, May 31, 2015, 3:43 AM UTC

vitovito wrote: post #9

chachalarue wrote: post #7

Earlier quoted context omitted.

Any recommendations for converting books into another format besides PDF? Ever since I read SICP as a texinfo in Emacs while working on another screen I've been looking for an easier automated way to convert my library to texinfo or LaTeX source.

If by "automated" you mean "cheap/free," no. If by "automated" you mean "I don't have to do any work" irrespective of cost, yes. When you "scan" a book, you're taking photos of the pages, which you can then run through OCR, but OCR, even of a page scanned with a flatbed scanner, is not going to understand page layout and a variety of typefaces perfectly. You're going to have a lot of errors to correct, and usually so…

If the book is just text and headings, can't you expect OCR to do a good job without any real need of human intervention?

Re: How can I legally convert a “dead-tree” library (1050 books) to eBooks?

#13

Post by qubex » Sun, May 31, 2015, 3:48 AM UTC

Thanks to everybody who took the time to answer. I'll look into the various proposed solutions. Thanks.

Re: How can I legally convert a “dead-tree” library (1050 books) to eBooks?

#14

Post by vitovito » Sun, May 31, 2015, 4:01 AM UTC

wodenokoto wrote: post #12

vitovito wrote: post #9

Earlier quoted context omitted.

If by "automated" you mean "cheap/free," no. If by "automated" you mean "I don't have to do any work" irrespective of cost, yes. When you "scan" a book, you're taking photos of the pages, which you can then run through OCR, but OCR, even of a page scanned with a flatbed scanner, is not going to understand page layout and a variety of typefaces perfectly. You're going to have a lot of errors to correct, and usually so…

If the book is just text and headings, can't you expect OCR to do a good job without any real need of human intervention?

Depends on what you mean by a "good job" and what you're doing with the results. 80-90% recognition isn't good enough if it means when you search for a term it doesn't show up because the OCR saw "rn" and wrote "m", or if you're having it translated, or having it read aloud with a text-to-speech synthesizer.

In my tests, we were seeing accuracy problems of 10-15% of lines needing correction, and this is a book that was primarily headers and text.

Sometimes this is character-level issues, like I cited above.

Sometimes this is dust, debris, shadows or markings being confused with text.

You get a little closer by running spelling and context checks against the words, but it's never 100% accurate. And if you aren't looking at the original pages, or you need automated systems to search/parse/translate/etc. the text, you need it to be 100% accurate, which means you need a human editor.

OCR isn't a solved problem.

Re: How can I legally convert a “dead-tree” library (1050 books) to eBooks?

#15

Post by wodenokoto » Sun, May 31, 2015, 11:46 AM UTC

vitovito wrote: post #14

wodenokoto wrote: post #12

Earlier quoted context omitted.

If the book is just text and headings, can't you expect OCR to do a good job without any real need of human intervention?

Depends on what you mean by a "good job" and what you're doing with the results. 80-90% recognition isn't good enough if it means when you search for a term it doesn't show up because the OCR saw "rn" and wrote "m", or if you're having it translated, or having it read aloud with a text-to-speech synthesizer. In my tests, we were seeing accuracy problems of 10-15% of lines needing correction, and this is a book that w…

Thank you for the detailed response. I honestly thought it was a solved problem (as in better performing than a human) as long as there is just running text.