Wednesday, 18 February 2015

Facebook Will Soon Be Able to ID You in Any Photo


Face
Facebook’s DeepFace System

When one appears in a photo taken at any event, they are recognized most of the time but a machine may fail to do so unless a computer has been programmed to look for you and has been trained on several photos of the face with high quality images to examine you.

However, in Facebook, where one will find the largest collection of personal photographs, the technology is making some headway in that direction. The California based company besides several other corporate players in the field – Facebook’s DeepFace System, is presently as accurate as a human being with few constrained facial recognition tasks. According to Yann LeCun, a computer scientist at New York University in New York City directing Facebook’s artificial intelligence research, states that the purpose is not to intrude on the privacy of Facebook user which is more than 1.3 billion active users but to protect it.

When DeepFace tends to identify a face in one of the 400 million new photos which the users upload daily, ‘one will get an alert from Facebook informing them of their appearance in the picture. The user then has the option to blur their image from the picture in order to protect their privacy.

Automated Facial Recognition – Usage/Limitation - Unknown

Most of the people tend to get disturbed on being identified particularly in stranger’s pictures and Facebook has already begun using the system though its face tagging system reveals only the identities of their friends. Besides DeepFace, the U.S government has also been funding in university based facial recognition research while others in the private sectors, like Google together with other companies are proceeding with their own projects to identify individuals automatically who tend to appear in videos and photos.

How automated facial recognition will be used and its limitation on the law is unknown though once the technology takes shape it would create many privacy problems as it solves. Brian Mennecke, an information systems research at Iowa State University in Ames, studying privacy states `the genie is or soon will be out of the bottle and there will be no going back’.

Social Face Classification 

LeCun comments that identifying a face could be much harder problem than detecting it which unlike fingerprints constantly tends to change. By just a smile, the face gets transformed where the corners of the eyes wrinkle, nostrils flare and the teeth are seen. When the head is thrown back with laughter, the apparent shape of the face contorts and when one has the same expression, the hair varies from each photo particularly after a visit to a hair dresser.

Yet most of them are recognized without any effort even if they have been seen in just one photo. The greatest advantage of DeepFace and the aspect of the project which sparked off is its training data. The DeepFace paper mentions the presence of a data set called SFC – Social Face Classification, which is a library of 4.4 million labelled faces taken from the Facebook pages of 4030 users and though users have given permission to Facebook to use their personal data on signing up for the website, DeepFace research paper does not make a mention on the consent of the owner’s photos.

No comments:

Post a comment

Note: only a member of this blog may post a comment.