Optical Character Recognition Still Struggling With Handwriting

Ian Lamont recently asked Google if they planned to extend their transcription of books and other printed media to include public records, many of which were handwritten before word processors became ubiquitous. Google wouldn't talk about any potential plans, but Lamont found out a bit more about the limits of optical character recognition in the process: "Even though some CAPTCHA schemes have been cracked in the past year, a far more difficult challenge lies in using software to recognize handwritten text. Optical character recognition has been used for years to convert printed documents into text data, but the enormous variation in handwriting styles has thwarted large-scale OCR imports of handwritten public documents and historical records. Ancestry.com took a surprising approach to digitizing and converting all publicly released US census records from 1790 to 1930: It contracted the job to Chinese firms whose staff manually transcribed the names and other information. The Chinese staff are specially trained to read the cursive and other handwriting styles from digitized paper records and microfilm. The task is ongoing with other handwritten records, at a cost of approximately $10 million per year, the company's CEO says."

Beat up Martin = Eat up Martha

Beat up Martin = Eat up Martha

Re:Fortunately, there is an alternative

i've outsourced all of my computer applications and software needs to India.

instead of using PowerPoint at meetings, i just have two Indian women in bikinis hold up large displays with my bullet points written on them--they even do slide transitions.

instead of an e-mail client, i use an Indian courier. it takes a while for me to communicate with international clients, but i receive practically no spam.

and rather than a word processor i have a guy with a notepad that a dictate to. he also offers me helpful tips when he notices that i'm trying to write a letter.

then there's the 17-year-old i have doing my taxes. i don't even think he's out of high school yet, but he beats Turbo Tax any day.

but you should really see the guy i have simulating Windows Vista for me. he wears this really slick suit, moves really slow, and everyone once in a while he comes up to me and kicks me in the balls.

Translation server error

Now that's a name that'll be remembered [boingboing.net]

Better approach?

It seems to me that it would be better to OCR everything and contract the proof-reading to the Chinese firm. The wide variation of writing styles and letter forms may make 100% accuracy of OCR impossible for this task, but starting from OCR should reduce the task, shouldn't it?

Re:Better approach?

Chinese proof-reading? Only if you want your documents in Engrish.

Re:Better approach?

No.

I own a microfilm digitization / OCR shop. We work with tons of old records such as the ones referenced in this story, as well as old HR docs, check stubs, time cards, architectural drawings, you name it. If you OCR cursive, you don't get back 80%, or 70%, or even 30% accuracy . . . you get back a bunch of pseudo-random (to our eyes) characters which are in NO WAY related to what the actual text is. About the only handwriting recognizable using today's tech is block-print, like you find on engineering diagrams. The technique in this article is pretty standard operating procedure, and has been for some time -- much easier to put a few hundred people on the project and grind through it (and cheaper too compared to data entry rates here in the US -- about 1/3 the price). That usually includes double-keying to check everything and a 99.99999% accuracy guarantee.

Just FYI, there are only a few OCR engines out there. Probably the most commonly used is the ABBYY engine, which is both OEMed and sold directly as desktop- and server-based products by ABBYY. There are a few others as well, and despite their differences, most have pretty much the same capabilities and accuracy. But OCR of cursive, especially of the docs cited in the article where you don't have someone sit down and "train" the machine first with handwriting samples, is still one of the great "unsolved" computing problems. I expect we'll have the capability in the next decade or so as processor core density, memory, and storage continues to increase at their current rate -- eventually, the machine will be able to "brute-force" through the docs just like the Chinese data entry folks in this article.

Re:Better approach?

I expect we'll have the capability in the next decade or so as processor core density, memory, and storage continues to increase at their current rate -- eventually, the machine will be able to "brute-force" through the docs just like the Chinese data entry folks in this article.

In the next decade or so we will have increased our processing power about 1000 times over. This work is scalable "sideways" — two pages can be processed by two computers independently. Which means, a thousand of today's computers could've done the work @home-style.

The problem is not with the processing power — it is the lack of algorithms. You and I reassemble the hand-written characters quite differently from how today's computers do it. The software will need to be created — and it is not the lack of CPU/memory/storage power, that's holding it.

One thing for sure is that the new algorithms will need to use the spell-checking engine(s) to better guess, what the next letter might be. On top of that, they would need to be equipped with grammar-checkers too, to be able to guess the next word, however illegible. Human speech (and thus writing) is quite redundant often — even if a misplaced coma can reverse the meaning on occasion.

Our brain certainly uses its knowledge of both the general rules of the language and that of the domain of what's written — this is why another doctor can decipher another doctor's handwriting, for example, that's infamously illegible to mere mortals. The software will have to do the same — and it can start doing it already.

parts of the problem are solved

The US Post Office has, for years, had fairly reliable automated reading of handwritten digits, which is used to auto-sort and -route mail by zipcode. It can handle some pretty terrible handwriting, crazy arrangement on the envelope, and unlikely variations, so only a relatively small percentage of letters are spit out to be read by human eyes.

Its task is made easier by the fact that they're locating and segmenting fixed-length sequences that are usually at least somewhat separated: they're looking for eith

Re:

Couple of points...

1) MS Research probably has some of the best work being done on handwriting recognition, including imaged documents. However it is no where near the needed levels. Google would be better off to work with Microsoft on stuff like this, than the motto of screw anything MS is doing and we will recreate it ourselves.

2) On your Vista Tablet PC, the reason you can get 90-99% levels of recognition is that TabletPCs and Vista/Windows use a concept called 'ink' (that goes back to early work at Mic

Re:

Teaching someone English at that level would be more difficult that teaching them to recognize characters. In ancient Rome the people who engraved dies for coins weren't always literate, but they managed for the most part to get the inscriptions right. Barbarians who made copies had more trouble, but then perhaps they thought the inscription part was purely decorative allowing for artistic interpretation. Or perhaps they weren't flogged for making mistakes. Point is, you can copy without having the high

Re:

Doesn't this suggest an obvious solution to CAPTCHA? Just use cursive text rather than try to obscure the text with funky backgrounds. If the spammers do manage to crack the CAPTCHA, then incorporate their technology into mainstream OCR programs.

Half the time..

I can't even read people's handwriting, I hardly expect a computer to.

Re:Half the time..

Hell, I can't even read my own handwriting. Yeah, this is probably not going to happen.

Re:

I've been using a computer since I was a kid, 25 odd years now. I can't write. I don't believe I ever really learned it.

I can print if I have to, though I usually ask my wife to do it because my hand gets sore after filling out a one page form. (In contrast I can easily type for 14+ hours at a stretch.)

I guess I get the point of handwriting recognition, for historical documents, but do we really need it for future devices?

general OCR harder than CAPTCHA OCR

There is a simple reason that general OCR is much harder than cracking a CAPTCHA. General OCR has to recognize text *reliably*. CAPTCHA breakers are thrilled with a 10% success rate, because they use distributed systems created by worms to do the hard work a million times over. If you got 10% of the words right when scanning historical records you might as well not bother.

Too variable, less reference

An OCR program can include a bank of fonts, and even when there is some sort of spill/ink blot/whatever on the paper, it has a solid reference. Handwriting isn't so easy, because humans don't always write their "Q"s with the line in the exact same spot and other fluctuations. Even if you gave a computer a point of reference (neatly drawn letters corresponding with their actual alphabetical values), a computer probably couldn't get it for a lot of people with inconsistent handwriting.

Now, with context and improved technology, I don't think that handwriting recognition is impossible. I have a feeling that it will be a technology like speech recognition: never perfect, and it will require training.

Re:Too variable, less reference

While handheld technology is indeed getting better, it's not directly applicable to the problem at hand. Real-time handwriting analysis uses stroke analysis as well as shape analysis to determine the letter(s). That is, the order in which you construct your letters matters very much. For example, if you crossed your T before drawing the vertical bar, the engine may have a difficult time figuring out what you intended.

When OCRing documents, all of that 'meta-information' is lost.

Now you have a training dataset.

Now you take the human translated recognition, and use it to train your genetic algo or neural net against the original images.

It's more complex problem than people imagine

There is an on-line archive of all people that have passed trough Ellis Island (http://www.ellisisland.org/search/passSearch.asp). It consists of retyped (OCR-ed?) ship manifests. Manifests are lists of passengers, with names, places of births and similar information. In original, they are written by hand, in cursive scripts (as expected for late 19th and early 20th century).

Problem is not with the script, but with appropriate context. Someone who retyped this, did not know what to expect in these forms.

My grand-grand father's place of origin was written as "Lipovqani, Slovenia". Pair "lj" was recognized as "q". For someone who is native English speaker "lj" one next to other does not make too much sense. But for anyone with Slavic origin, "q" does not make sense (it's only in foreign words), and "lj" does make sense since it is a way to write "soft l" voice like in "Richelieu".

Ok, maybe that was not the an easy part to guess. But "Slovenia" was serious error. In that moment, Slovenia did not exist. It was part of the Austro-Hungary, and it did not exist as single entity inside it. What was really written was actually "Slavonia". That's an area in Eastern Croatia, and it *was* an entity inside Austro-Hungary.

Should I mention that I was not able to track my grand-grand mother and my other grand-grand father?

Pretend its a string?

Can OCR properly trace the lines at least to replicate it? Meaning, it could make a vector replica of the handwriting? Would be neat if it could do that, then try to straighten out the lines, perhaps to simulate the possible path the original writer took to write it. Of course, the software will have to figure out intersections. Maybe a path of logic would be to know what turns a handwriter would NOT take, and then determine individual letters from that.

Combine that with other logic, like finding "dots" would indicate an i or a j, and maybe it will improve.

Re:Use them as CAPTCHA...

Joking or not, that's kind of the idea behind reCAPTCHA. It takes words that OCR failed on and uses them as CAPTCHAs. The same idea could work for handwriting. http://en.wikipedia.org/wiki/ReCAPTCHA [wikipedia.org]

Re:Optical Character Recognition is the Correct Te

For a moment there, I was picturing some new technology that could distinguish between C, PERL and and Java written on scratch paper.

In pseudocode:

IF LooksLikeC THEN "This must be C code"
IF LooksLikeJava THEN "This must be Java code"
// undecipherable
ELSE "Must be Perl code"

Re:New strategy

You joke, but there really in very little reason to teach children handwriting/script/cursive (whichever you want to call it). The point of cursive was to speed up writing. It was never any good for readability. In today's world, if you need to write a lot of stuff, you are generally going to type it on a computer. Since just about anything that we would want to write by hand will be short, the speed gain would be minimal. Thus spending time and resource to teach every kid to write a useless, illegibly font is pretty pointless.

Re:New strategy

Except for most of us it's faster to write with your hands.

Writing by hand, you can jump letters and make abbrevs, you can draw diagrams right in there, and not to mention it feels a lot better. I don't know why but sitting and typing on my computer, and same when I used to paint minis, feels painful and stuffy. With the option of either typing or writing I'd definately take writing. Sure, with typing on a computer you can erase stuff quickly, but text editors have always been shitty for me (stuff like AbiWord often having graphical glitches or plain slow, text editors too or just lame feeling) and hitting a bunch of blocks to make words does not feel as good as actually writing down the words.

I never mastered cursive properly. I write "script", but write while skipping letters in my notes and using small symbols (batman symbol, drawn as a W in a circle, for example, is distress; three points is "donc", ds dans, etc and it changes depending on context). I write fairly fast, and imho much faster than when I type, if only because when I type I often hit the wrong keys; often being once a paragraph, and it's often because I can't get my mind straight on the keymap, or my fingers hit in the wrong order.

lecture notes

Have you ever taken lecture notes?