Camera phone helps label snaps
* 23 December 2005
* NewScientist.com news service
KNOWING where you are, what time it is and who you are with is obviously a huge help when it comes to filing a photograph in your collection. It also happens to be information that can now be compiled by any Bluetooth-enabled camera cellphone.
The phone will allow the growing number of camera phone users to organise their digital photo albums by automatically identifying and labelling the people and places within each snap, as they are taken.
The concept, being developed by Marc Davis of Yahoo’s Berkeley research lab in California, is based on a central server that registers details sent by the phone when the photo is taken. These include the nearest cellphone mast, the strength of the call signal and the time the photo was taken.
The system also identifies the other Bluetooth-enabled cellphones within range of the photographer and combines this with the time and place information to create a shortlist of people who might be in the picture. This can then be combined with facial-recognition algorithms to identify the subjects from the shortlist.
Facial recognition software on its own can only identify people with 43 per cent accuracy from the grainy shots taken by camera phones, but in tests Davis and his team found that by combining it with context information the system could correctly identify people 60 per cent of the time. The context information can also be combined with image-recognition software to identify places within photos.
Related Articles
* Cell phone could identify its owner by their walk
* http://www.newscientist.com/article.ns?id=dn8161
* 14 October 2005
* Camera phones will be high-precision scanners
* http://www.newscientist.com/article.ns?id=dn7998
* 14 September 2005
* Illicit snappers caught infrared handed
* http://www.newscientist.com/article.ns?id=mg18725156.500
* 03 September 2005
Weblinks
* Marc Davis, Yahoo Research Berkeley
* http://research.yahoo.com/researcher/marcd.shtml
* Yahoo Research Berkeley
* http://research.yahoo.com/berkeley/
* Bluetooth
* http://www.bluetooth.com/
