best HW + SW to convert books to epub

[rumpumpel1]

Hi,

what's currently the best Hardware and Software to scan books and to convert them to epub with little effort ?

The books have no fancy layout, no images, just plain text and the books may be taken apart.

[HarryT]

Moved to the "Workshop" forum.

[HarryT]

If you really mean "best" as in "price is not a consideration", then you should be looking at a device such as this:

http://www.imageaccess.de/?page=Scan...V2Professional

This is the type of device that professional scanning bureaux use.

If price is a consideration, and you can destroy the books, then a scanner with an automated sheet feeder is probably what you want to be looking at. Something like this:

http://www.amazon.co.uk/Fujitsu-Scan.../dp/B001VGJ7JM

[Toxaris]

Originally Posted by rumpumpel1

to convert them to epub with little effort ?

No can do. It is either fast and sloppy or slow and thorough. (A lot of) Effort is required.

[HarryT]

Originally Posted by Toxaris

No can do. It is either fast and sloppy or slow and thorough. (A lot of) Effort is required.

A decent OCR program like Abbyy FineReader will give you pretty good results (as in perhaps an error per page), and that may be acceptable for casual reading purposes, but you are of course right in saying that proof-reading is essential if you want an error-free book.

[rumpumpel1]

that's amazing: even for simple layouts and plain text there is a remaining error rate of one error per page ? What kind of errors are these ? Can they be corrected with a spell checker of a decent office program?

[HarryT]

Originally Posted by rumpumpel1

that's amazing: even for simple layouts and plain text there is a remaining error rate of one error per page ? What kind of errors are these ? Can they be corrected with a spell checker of a decent office program?

These are "normal" OCR errors, where the shapes of letters look similar, eg "clock" instead of "dock" ("cl" and "d" are very difficult for OCR to tell apart). A spell-checker won't help, because they are real words - just not the right word.

A decent OCR program has an accuracy rate of better than 99.9%, but a typical page has around 2000 characters on it, so that means about 2 character errors per page. Some of these the OCR program's spell-checker will fix for you, but some it will get wrong.

[Toxaris]

It really depends on the source. Some errors can be found by a simple spell checker, but some not of course. That is one of the reasons I started to create my tools to remove a lot of OCR errors. It can be spelling, but also punctuation that is going wrong. Not to mention styling an others. A lot cannot be found with the standard tools.
The better the source (and scan), the better the results of the OCR program. ABBYY is doing a good job is the scan is good.

[dwig]

Originally Posted by Toxaris

It really depends on the source. Some errors can be found by a simple spell checker, but some not of course. ...

Also, the spell checker can't be relied on to auto-correct with more than only a moderately decent accuracy which means manual correction of each found error.

Also, the error rate for OCR software drops significantly when working from scans of books in poor condition (e.g. foxing, stains, yellowing, ...), printed poorly, and/or printed on poor quality paper.

[cromag]

Also, slight imperfections in the paper -- dark spots, a stray fiber, etc. -- can be mistaken for punctuation marks like periods and commas.

[Hitch]

Originally Posted by HarryT

These are "normal" OCR errors, where the shapes of letters look similar, eg "clock" instead of "dock" ("cl" and "d" are very difficult for OCR to tell apart). A spell-checker won't help, because they are real words - just not the right word.

A decent OCR program has an accuracy rate of better than 99.9%, but a typical page has around 2000 characters on it, so that means about 2 character errors per page. Some of these the OCR program's spell-checker will fix for you, but some it will get wrong.

Exactly. Given how well-used Abbyy is, it's almost ALWAYS things like hat for fiat, and the like. No spellchecker will find those. And formatting errors? A BOATLOAD more than 1-2 per page. Almost all the "work" is in fixing the formatting, cleaning up the text, removing spans, and all that.

This part, from rumpumple1:

...and to convert them to epub with little effort

Simply doesn't exist. There's no such thing. The only material that can be "convert[ed] to epub with little effort" are those files that are clean to begin with--which means, what you've done between the Scan/OCR and the time you start to actually make the ePUB.

Hitch

[GrannyGrump]

Unless, of course, you forego OCR entirely, and simply go with images of the printed page only. Of course, that way you lose the capability of reflow, search, annotating, all the things you can do with text that are not possible with images. A mighty gloomy result, I think. (There are quite a number of epubs out in the wild that are made like this, recognizable before reading by their HUGE file size. They are really like a pdf in disguise.)

[Ghitulescu]

Originally Posted by rumpumpel1

that's amazing: even for simple layouts and plain text there is a remaining error rate of one error per page ? What kind of errors are these ? Can they be corrected with a spell checker of a decent office program?

Unfortunately, you HAVE TO WORK, unless you're using low standards of quality.
Since most people are different, they have different needs and perceptions, and therefore there are zillions of TVs, computers, cars, etc. because every single human wants a feature more than another feature (like colour red for cars ). Therefore an automatic procedure, designed to fit statistically all customers won't satisfy all.

[GrannyGrump]

rumpumpel1 queried:

What kind of errors are these ? Can they be corrected with a spell checker of a decent office program?

A spell-checker can't help when the OCR interprets a dirt-spot as a colon. It can not help when the OCR gives a legitimately spelled word that is the *wrong* word, as in Harry's example of "clock" versus "dock". Other frequent OCR errors include "die" for "the", digit 1 instead of lower-case L or upper-case I (or any other mis-match combination of these three characters), the letters "rn" (that is lower-case R N ) turn out as "m" (that is lower-case M), etc, etc. Italics add more spice to the stew.

A *grammar-checker*, such as Microsoft Word provides, can help to some extent when it recognizes that a word is blatantly wrong for the containing sentence. Unfortunately, it is far from perfect still; and mostly concentrates on punctuation errors.

To get good results, you will have to physically proof-read the scan results, comparing against the printed page.

[HarryT]

Originally Posted by GrannyGrump

the letters "rn" (that is lower-case R N ) turn out as "m" (that is lower-case M), etc, etc.

An example of this is that I'm currently proof-reading the "J G Reeder" detective stories of Edgar Wallace for the MR library, and a speech affectation of his is to use the word "um" a lot to indicate pauses in his speech. On at least half the occasions in the PG text I'm proofing from, "um" is spelt "urn".