Search Engines for Handwritten Documents 172
An anonymous reader writes "Researchers at the University of Massachusetts have created a tool for automatically searching handwritten historical documents, such as the 140,000 pages that make up George Washington's personal papers in the Library of Congress. The most interesting part is that the papers are scanned versions of the originals and the search tool actually recognizes the handwritten text from these images."
Re:The search tool? (Score:2, Interesting)
Hard to read! (Score:3, Interesting)
Accuracy? (Score:2, Interesting)
Handwriting sucks (Score:4, Interesting)
I hate reading/producing anything longer than a post-it note that's in handwriting.
Re:Who still reads those? (Score:5, Interesting)
!WOW (Score:1, Interesting)
Useful for more than just historians (Score:5, Interesting)
Re:A waste? (Score:3, Interesting)
Interesting, but limited (Score:3, Interesting)
If you can put the queries in English, with the search engine taking care of translation, it would be even better. Then, extended historical study comes within everyone's reach and the classical studies (or humaniora) might be transformed.
Re:Who still reads those? (Score:2, Interesting)
Re:Who still reads those? (Score:5, Interesting)
The best part is I don't have to worry about backing up my lab books. The only real threat is fire, and it is no more dangerous than it is to CDs or hard drives.
While the cursive handwriting of the 1700's and early 1800's may seem curious to us (notably, the tall 's' that looks like an 'f'), it is a very easy style that is neat, legible, and painless. Notice how there are very few back strokes.
For those who are wondering, cursive is what you use when you get sick of trying to write in print legibly and quickly without getting carpal tunnel. Every culture has it. It's unfortunate it isn't common knowledge anymore in the US. Handwriting is a wonderful skill. It used to be people would judge others based on their handwriting skills in addition to their oratory.
More like twenty years ago ;-) (Score:4, Interesting)
I worked on an OCR system about 20 years ago. No pre-defined bitmaps of text, you trained the system on the font to be recognized. After a few hours you could turn it loose and it did fairly well. While goofing off we tried handwritten text. With good penmanship it worked to a degree.
Re:Who still reads those? (Score:3, Interesting)
Anyway, I watch Full Metal Jacket and it reminds me of Catholic school. To continue my rabmle, how many people who actually went to catholic school aren't curretly aethiest? I'm guessing not too many.
Here...let me help you out, mods:
(-1 Offtopic)
One important thing to understand about this... (Score:3, Interesting)
For instance, the software may be unable to distinguish the word bug from dog in one person's handwriting, but can still mark it with probabilities of the word's possible meanings.
If a person later searches for the word bug or dog at a future date along with other terms, a mathematical calculation can be done for the likelyhood of the match and the searcher can make his/her own judgement to the meaning of the text.
---
Conrad Barski
Re:One important thing to understand about this... (Score:2, Interesting)
In the legal field, finding context in a search is typically as (or more) important as finding a single word... Products like Summation (Summation.com) and Adobe's industrial strength Acrobat Capture (? - may have a new name... Server-based - uses "hot folders" that are monitored, batches, etc.) have OCR capabilities that are pretty flexible, reading from text, pdf, MS Word, JPEG, BMP, GIF, or TIFF... Of course, these can be expensive...
But, being able to get quickly to a target word is very useful indeed when the verbatim answer requires human eyes to confirm or contextualize, or
I used Acrobat Capture and a digital camera (I was not permitted to flatbed or sheetfeed scan - the items were deemed too "fragile") to make archive materals text-searchable for a law firm's special project, to very good result.
Granted, these materals were written in decent fonts, not handwritten, but with many graphic illustrations interspersed in them, which can trip up some OCR solutions. Capture could have read the documents' Japanese too, if I'd bought the correct Adobe plugin, and the project had required it.
Massaged correctly, OCR's come a long way, baby...