The Economist did a piece last month on paperless offices, or rather about one that decided to do away with its paper for good and is "now about halfway there."

Hopefully they follow this story until it ends, for I sure would like to be there when Breedlove & Associates crosses the finish line - if only to give them a little hand-written 'Congratulations' note with the hope that they'll respect my regards enough to not throw away that piece of paper immediately.

The author cites plenty of examples of people shinning paper for woodfree technology - Demand for office paper is now declining unlike a decade ago, the world's largest paper maker has closed 5 factories in America in this period, and students now take their laptops instead of printed notes to their lectures.

These are all major developments that bring efficiency and great promise for an easier life, but they've got less to do with saying au revoir to paper forever, and more about technology now pervading territories that were formerly untouched.

Emails, e-tickets, e-returns: these are all important trends, but they are probably asymptotic. For there are things we still do where paper won't budge, at least not just yet - Google recipes are good, but I still wouldn't bring my laptop to my kitchen; keyboards and pasta sauce don't taste very good.

Economist claims that "information thus appears to be becoming paperless roughly as transport has become horseless." The analogy is certainly a telling reflection of where we're headed. In our homes and offices however, I still think we'll be riding these horses for some years to come.

OCR can be useful in unexpected situations. When I was in University, old examination papers had been made available online as PDFs. These had been scanned and converted to the pdf format and were not text-searchable. I managed to convert them to searchable pdf using ABBYY - however, people who just want a few pages/ images OCRed to avoid typing might find commercial OCR software expensive.

There's Tesseract, the free OCR engine that Google is currently developing. We have built some applications using Tesseract as the back-end and have found that the accuracy of text recognition is not comparable to other commercially available solutions. Tesseract does not have a user interface making it an unattractive choice for someone who just wants a couple of pages OCRed.

Two other open source OCR software that we haven't used in-house are OCRAD and GOCR. Both these are command line programs requiring a separate front end installation for users who are not programmers. Their accuracy in character recognition has been reported to be lower than that of Tesseract.

Microsoft Office comes pre-packaged with Microsoft Office Document Imaging that is included in Office tools. The instructions of how to scan and digitize your documents using Office have been documented here. Though the OCR works reasonably well on screen shots, it fails to preserve page layout information. A document with two columns (e.g. scientific papers) produced an OCR output where every horizontal row was treated as a line. This effectively meant that the resulting block of text made very little sense semantically.

OCR Terminal was built to provide the user a free and easy way of performing OCR without having to install new software or having to pay for it. Automatic layout extraction and segmentation of image areas ensure that the page layout and formatting of the document is preserved accurately. We are currently working on a desktop client that will allow users to upload the document by dragging and dropping all the tiffs and pdfs into the client and get back perfectly formatted Word files.

And, OCR Terminal is free.

After decades of research in OCR technologies, machines today are much better today at reading text than their ancestors were. A 100% accuracy rate on some printed documents is not surprising today. Yet, there isn't one software that claims to be completely accurate (including us!), even within a defined data set, say clean white sheets with black 12pt printed text.

This problem has as much to do with technology as it has with liability, especially in error-sensitive settings. A secretary that makes an error in typing a document can be reprimanded or even fired. Shouting at a computer however is never a solution.

So if we may be so adventurous so as to venture a prediction, here's one: No matter how good OCR gets, it will never replace manual systems until legal jurisdictions expand to hold machines liable for the errors they make. Until that happens, no matter how intelligent we make our algorithms, we'll never be able to promise you can trust them blindly.

Of course that's no excuse - OCR Terminal remains the easiest and the most accurate OCR solution on the web, and we'll make sure it always does.