DJVU to ePub best results?

[Begemot]

What methods have you used to convert DJVU to ePub?

The current method I am using is as follows:
Open DJVU with DjVuLibre DjView 4.4
Export as PDF
Then add PDF to Calibre Library and make ePub(this is basically running pdf2html utility)

Problem is the conversion to PDF step.
4MB DJVU explodes to 35MB PDF!
Then PDF to ePub goes down to 20-25MB, but the results are less than stellar.
Slightly smaller problem is that running pdf2html on a 35MB PDF takes about a half an hour, but I could live with that if the quality was good.

All of this was done on Ubuntu 9.10, but I would be interested in hearing about DJVU to ePub solutions on Windows or Mac as well.

[charleski]

DjVu doesn't contain text, it works on images, and that's your problem. It uses an image-compression technology that's highly optimised for text and allows far smaller file sizes than other formats that target more general image types.

You could export the images and OCR them (lots of work to catch the errors), or you could try slapping the images all together as-is (which is what you describe above, possibly with some down-ressing to make it look even worse). Go back to the author and get a file that contains text, because DjVu is useless for your purpose.

[Websterny]

I know nothing about this topic, but I have the same problem, and I note that DJVU files do seem to contain text - that is, the documents are searchable. They can be converted to non-searchable PDF with the print command (assuming you have a PDF print driver installed). But this significantly detracts from the utility of the files. And their size is a multiple of the original DJVU. There has got to be a better way.

[BobC]

Originally Posted by charleski

DjVu doesn't contain text, it works on images, and that's your problem. It uses an image-compression technology that's highly optimised for text and allows far smaller file sizes than other formats that target more general image types.

DJVU's Can contain a hidden text layer (which is used in the search feature). This layer can be extracted and used as the basis for any other conversion.

For example most of the DJVU files on The Internet Archive (TIA) contain such a layer and I have used them as a basis for FB2 books.

Of course the files the OP is working on may not have such a layer as the original text may not have been OCRd and associated with the image layer.

BobC

[pholy]

BobC - Can you tell us how to extract that hidden text layer? I haven't run across any DJVU books that I recall, but it would be good to know how to convert them when possible.

[BobC]

Originally Posted by pholy

BobC - Can you tell us how to extract that hidden text layer?

Either Highlight the text in the Image view and use <CTL>C to copy it to the clipboard or use the "Export Text" feature in WinDJView or some similar Viewer.

BobC

[Begemot]

OP here, I resorted to using export Text in WinDJView.

This gets you a text dump with no formatting whatsoever. For my Libre it works well enough, but in general, this procedure is suboptimal.

Most DJVU files do seem to have a text layer (unless there is some on the fly OCR happening when you select an area on the page, which seems unlikely).

Thus, there must be a way(at least theoretically until someone writes a converter) to preserve the formatting in the text layer.

[bugmen0t]

I just print from any djvu reader into a pdf printer, like primopdf or other... only thing is: 240 pages book, 4.5Mb originally was transformed into 40Mb... maybe trimming the quality of the pdf down...

[BobC]

Originally Posted by Begemot

OP here, I resorted to using export Text in WinDJView.

This gets you a text dump with no formatting whatsoever. For my Libre it works well enough, but in general, this procedure is suboptimal.

Most DJVU files do seem to have a text layer (unless there is some on the fly OCR happening when you select an area on the page, which seems unlikely).

Thus, there must be a way(at least theoretically until someone writes a converter) to preserve the formatting in the text layer.

I can assure you that the text layer is just that - text; it's purpose is simply to provide the search capability. There is no formatting and in many books there are OCR "mis-reads".

If you want to understand DJVUs then you need to get the spec and study it. I've done quite a bit of work with adding TOCs to existing DJVUs and have converted a couple of books to FB2 - this involves manually proof-reading and correcting the dumped text then formatting it to match the original (italics, bold etc).

Don't expect too much out of what is a by-product of the search function.

BobC

[DaleDe]

There is a description of DJVU in the wiki.

[Mex5150]

OK, resurrecting an old thread, sorry. But, I'm now stuck with this problem, has conversion moved on since this was originally posted? I can extract the text in the djvu file (in a few different ways) but I can't get inline images to convert. What I want is to go from djvu to epub, I'm not worried about extra steps in-between as long as both text and images make it to the final epub. Like the OP I'm running Linux, but I can probably borrow a windows machine for a while if that's the only way to do this. Any ideas?

[Doitsu]

Originally Posted by Mex5150

Any ideas?

Calibre has a built-in DJVU converter that allows you to convert DJVU files with text layers to epubs.

[Mex5150]

Originally Posted by Doitsu

Calibre has a built-in DJVU converter that allows you to convert DJVU files with text layers to epubs.

Calibre was the first thing I tried, but that just converts the text without the images. (probably should have mentioned I'd already failed with Calibre, oops)

[Doitsu]

Originally Posted by Mex5150

Calibre was the first thing I tried, but that just converts the text without the images. (probably should have mentioned I'd already failed with Calibre, oops)

AFAIK, you can either extract the text layer or images, but not both. In order to keep the inline images, you'll have to convert the DJVU files to PDF files and then OCR them like any other PDF file with ABBYY Finereader.

If your Nook is rooted, you might be able to install KOReader, which supports DJVU files.

Alternatively, if the Nook PDF app doesn't have a reflow function, you might be able to use k2pdfopt to reformat PDF files generated from your DJVU files for your NOOK. (In case you can't figure out the optimal conversion settings, the author, willus, provides tech support in the k2pdfopt forum.)

[Mex5150]

Originally Posted by Doitsu

AFAIK, you can either extract the text layer or images, but not both.

Well, that's a pain. Looks like I'll have to survive without the images then.

Thanks for your help anyway.