March 19, 2005
Hitachi Develops Camera Phone Software That Reads Japanese
Hitachi Ltd. has created OCR (optical character recognition) software for camera phones that could open up a new range of services, according to an article in Forbes
"OCR programs that can read Japanese characters and convert them into text usually require tens of megabytes of memory. But Hitachi has developed an algorithm that requires less than 1MB, which is compact enough to run on a cell phone.
The software analyzes the shapes of the characters and the relative positions of words, converting the text images captured by the phone's camera into text data that can be transmitted over the Internet.
This setup can be used for services that provide users with additional information when they use their camera phones to photograph interesting articles in a magazine or catalog.
Services of this nature already exist based on 2-D bar codes purposely printed in magazines and catalogs, but Hitachi's new OCR software provides more versatility, including the ability to access information on products, restaurants and other topics seen in older publications. "
Related:
-- SpeechGear translates documents and street signs - The Office of Naval Research (ONR) thinks cameraphones can be put to good use serving as near-instant translators for documents and street signs.
"Take a picture of an Arabic newspaper, a Chinese menu or a Farsi instruction pack, and send the image to the servers of SpeechGear, an ONR-backed company based in Northfield, Minnesota.
SpeechGear processes the image, looking at changes in pixel color to see where the text is in the image. It then orients the picture around the text, adding pixels if needed to correct a skewed image. Once the picture is right, SpeechGear uses optical-character-recognition software and translation databases to see what's being said. An English version of the text is sent back to the phone in about 15 seconds.
The Permanent Link to this page is: http://www.textually.org/picturephoning/archives/2005/03/007590.htm
