2 ms·

Please show me the software to not only do OCR, but also that will reflow these what are essentially bitmap-representations of books (or, at best, 'PDF-like rep

by roel_v 14y ago

Please show me the software to not only do OCR, but also that will reflow these what are essentially bitmap-representations of books (or, at best, 'PDF-like representations') to a format such as epub which is required to get the full 'e-book experience'. For the record, I've tried my hand (and failed) on software to do just this. It's a Hard problem, Hard in the CS and engineering sense, and sufficiently hard that I don't see it happening reliably in the first years, and there is to the best of my knowledge no software available that even comes close. To anyone who is going to suggest Calibre, please don't, lest you make yourself look like you know nothing about the issues I'm talking about.

bornhuetter 14y ago
> To anyone who is going to suggest Calibre, please don't, lest you make yourself look like you know nothing about the issues I'm talking about. Can you please elaborate on this. I've never tried, but I thought Calibre could strip the DRM from kindle ebooks. Or do you mean that Calibre can't handle the general case of arbitrary DRM (via OCR methods)?
- roel_v 14y ago
  First, Calibre doesn't do OCR (at least not the version that I last looked at, maybe a year or so ago). But then still, even if it would, having a bunch of OCR'ed pages doesn't make an ebook. Any eventual conversion software would have to strip page header and footer material, recognize chapter titles and paragraphs, etc.; and for technical books it's a magnitude worse still - it would have to recognize sidebars, images, multi-level chapter headings, ... Converting a pdf into an ebook (by ebook I mean: epub or similar formats, I recognize that that is a fairly narrow definition of ebook; some people call scanned jpg's an 'ebook'), even without accounting for DRM or OCR, is Very Hard. PDF is only page layout, it doesn't have any semantics; ebooks are build around semantic markup (otherwise you can't make an automatic ToC, or reflow pages when the user increases font size, for example). My point is: there is no software (to the best of my knowledge, and I've looked) that can convert bitmaps into their equivalent 'ebooks'. So, it's not quite as easy (actually it's so much harder so as to approach impossible, save from manual conversion) as the GP would make it seem.
  - bornhuetter 14y ago
    Thanks for the clarification. I agree that converting a BMP to a properly formatted eBook is next to impossible without at least some manual intervention (or using an expert system based on manually constructed book templates). I do wonder how far Google has got with their internal software - and if they care at all about formatting, or just the text.
GFischer 14y ago
(Edit: from the other thread I understand roel_v's definition of an ebook includes "chapter titles and paragraphs" and all the formatting, not just a readable epub file which was my definition) I've done it, it was not a one-step process, the results were not pretty but it was readable. It's not a Hard problem (unless you mean Hard to do automatically and get good results, in which case I agree). I've got the tools at home, but I took the steps from the Mobileread forums. One possibility I just read there is, if you have a scanned PDF, PDF text extractor, then Use OpenOffice to save as epub. Also, you can just skip all that and just read the PDF from the ebook reader.

by roel_v 14y ago

bornhuetter 14y ago

> To anyone who is going to suggest Calibre, please don't, lest you make yourself look like you know nothing about the issues I'm talking about. Can you please elaborate on this. I've never tried, but I thought Calibre could strip the DRM from kindle ebooks. Or do you mean that Calibre can't handle the general case of arbitrary DRM (via OCR methods)?

roel_v 14y ago
First, Calibre doesn't do OCR (at least not the version that I last looked at, maybe a year or so ago). But then still, even if it would, having a bunch of OCR'ed pages doesn't make an ebook. Any eventual conversion software would have to strip page header and footer material, recognize chapter titles and paragraphs, etc.; and for technical books it's a magnitude worse still - it would have to recognize sidebars, images, multi-level chapter headings, ... Converting a pdf into an ebook (by ebook I mean: epub or similar formats, I recognize that that is a fairly narrow definition of ebook; some people call scanned jpg's an 'ebook'), even without accounting for DRM or OCR, is Very Hard. PDF is only page layout, it doesn't have any semantics; ebooks are build around semantic markup (otherwise you can't make an automatic ToC, or reflow pages when the user increases font size, for example). My point is: there is no software (to the best of my knowledge, and I've looked) that can convert bitmaps into their equivalent 'ebooks'. So, it's not quite as easy (actually it's so much harder so as to approach impossible, save from manual conversion) as the GP would make it seem.
- bornhuetter 14y ago
  Thanks for the clarification. I agree that converting a BMP to a properly formatted eBook is next to impossible without at least some manual intervention (or using an expert system based on manually constructed book templates). I do wonder how far Google has got with their internal software - and if they care at all about formatting, or just the text.

GFischer 14y ago

(Edit: from the other thread I understand roel_v's definition of an ebook includes "chapter titles and paragraphs" and all the formatting, not just a readable epub file which was my definition) I've done it, it was not a one-step process, the results were not pretty but it was readable. It's not a Hard problem (unless you mean Hard to do automatically and get good results, in which case I agree). I've got the tools at home, but I took the steps from the Mobileread forums. One possibility I just read there is, if you have a scanned PDF, PDF text extractor, then Use OpenOffice to save as epub. Also, you can just skip all that and just read the PDF from the ebook reader.