KBookOCR 2.0: Baseline OCR at Last
Off the Beat: Bruce Byfield's Blog
Every year or so, I need to convert a printed page to text. It's not oftener, because, while I do have a collected letters project that would benefit from optical character recognition (OCR), I only work on it sporadically. When I have to sign a publisher's agreement or some similar document, I can generally just send a scanned image.
But every year or so, only OCR will do for one need or other, and I plunge into a quick survey of the available free software tools. The results have always been fairly dismal and more trouble than they're worth. However, with the 2.0 release of KBookOCR, a base-level reliability and convenience is now available.
The truth is, OCR has been on of the weak points in free software for a long time.That hasn't stopped developers from trying, but the history of the free desktop is littered with OCR applications that have never reached their first general release. I suppose the reason is that most people use OCR even less frequently than I do, so that OCR is a low priority for distributions and has difficulty attracting large numbers of developers. The work of recognizing letters is finicky, and is second only to office suite filters in its lack of glamor.
Until now, the bright spots in OCR development have been few. Some years ago, Hewlett-Packard donated the code to Tesseract, which is now managed by Google, but, although Tesseract's character recognition from .tif images produces acceptably accurate plain text, its ability to retain formatting is relatively low. That means that it only does half of what is needed, since much of the convenience of OCR is lost if you have to do much fiddling with the format.
More recently, Cognitive Technologies released CuneiForm, which retains layout more accurately than Tesseract, but, like Tesseract, is a command line application. In other circumstances, I'm usually comfortable with command lineI applications, but dealing non-graphically with images always seems counter-intuitive and finicky. In my experience, it often leads to a couple of trials to get acceptable output -- and sometimes you waste your time scanning what the application is simply unable to read.
Enter KBookOCR 2.0
KBookOCR is a KDE front-end for CuneiForm. You upload a document from a PDF file or a scanned image, and thumbnails of its pages are presented in the left pane. In the middle of the window is a preview of the first page, whose size can be adjusted from the pane on the lower right, and the page on display from the bottom middle.
When you are ready to scan, you select the language and Layout OCR from the middle right pane, and the pages to convert from the upper right or left pane. Then you choose either to save output to file, or to open it in an editor -- on most systems, LibreOffice. It then scans at a rate of about twenty seconds per manuscript page, the progress being indicated by a bar in the middle right pane.
This is not the most organized layout I've seen; it would only make sense to have controls organized from left to right in the order that you need them, but this arrangement is only half carried out. Still, as illogical as the window's layout may be, it is far saner (or xsaner, I might say, since scanning is involved) than working from a command line, and knowing nothing about the results until you are finished.
Of course, the combination of KBookOCR and CuneiForm is still not perfect. As with any OCR, letter pairs that you might kern in a layout may be rendered wrongly. Complex layouts and decorative fonts -- especially scripts -- can also confuse the results.
All the same, so far as I'm aware, KBookOCR is the first free-licensed OCR that has a baseline accuracy and convenience that makes it less trouble to use than retyping a document from scratch.
A place to start
Having seen proprietary OCR applications, I'm well aware that KBookOCR is no match for them. For instance, I would appreciate KBookOCR more if it could work directly with font tables, and if I could identify formatting elements such as columns or graphics manually if necessary.
All the same, the latest release of KBookOCR is the first free-licensed OCR tool that I consider even marginally worth using. If it lacks some of the features I would prefer, in theory, at least, it should provide a reliable foundation for adding those features in later releases.
That's a minor victory that most people won't notice, but one worth celebrating all the same. In the latest KBookOCR, there's proof once again that it's not that free software is incapable of matching certain proprietary products -- it's just that in some areas, nobody's got around to doing the work yet. After all these years of watching for usable free OCR, I was starting to have my doubts.
Thanks for thisThank you for your article because much of it will help in creating new versions KBookOCR) If you are able to help, we would have been useful to your opinion. b0noI (my2you at ya.ru)
Richard Stallman calls for the W3C to remain independent of vendor interests.
The new release supports nine architectures, 73 human languages, and zero non-Free components.
Fedora developers release the first alpha version of Fedora 19, known as Schrödinger’s Cat, for general testing. The final release is expected in July 2013.
ack is a grep-like, command-line tool that has been optimized for programmers to search large trees of source code.
New features in SUSE Studio 1.3 include enhanced cloud integration, VM platform support, and lifecycle management.
The Linux Foundation recently announced that the Xen Project is becoming a Linux Foundation Collaborative Project.
Open source version of LiveCode is now available for developing apps, games, and utilities for all major platforms.
OpenDaylight is an open source software-defined networking project committed to furthering adoption of SDN and accelerating innovation in a vendor-neutral and open environment.
The new Gnome release includes privacy and sharing settings, allowing more user control over access to personal information.
Mozilla is collaborating with Samsung on a new web browser engine called Servo.