Facebook's new face recognizer that is 97.25% accurate!!

Facebook is out with a precise face recognition tool called the DeepFace which can recognize friends, family and the other associated people in a picture posted by user. The facial recognition software is apparently as sharp as the human eye according Facebook’s blog post and a report from MIT.

According to the reports, Facebook’s DeepFace analyzes if it’s the same face or a different one when compared and is not to be confused with the facial recognition which puts up a name on the person. The software is to take the auto tagging feature of the Facebook to all new level.

When asked about the faces of the two people with different looks being the same, the software gets it right at 97.25 percent compared to human’s level at 97.53 percent, according to MIT.   Facebook says  it is 25 percent more efficient than the present software which is in use, reports Business Insider.DeepFace recognizes faces in two steps.

First it corrects the image of the person to face forward, then it does a 3D model of a ‘normal forward facing human’ to fix the image, then it uses the deep face learning by simulating a neural network over the image and then a reoriented face is created. If the software finds some facial similarities between the two images then the software confirms it to be the same face.

      

On a study conducted by MIT and reported to Forbes, “The software is being tested with dataset of about 4 billion facial images belonging to around 4000 people which means each person had an average of thousand samples to be tested.”

Apparently, the facial recognition software has been around Facebook for quite a long time. But, with DeepFace, mistakenly tagging people in the pictures can be reduced while helping to tag a person in a low lit photo.

At present, Facebook’s DeepFace is only in the research form and by June this year, it will be presented at the IEEE conference in Ohio