Automatic face recognition has a large number of interesting applications. For example, Picassa and iPhoto use face recognition to help you tag people in your photo collections. Face recognition is also used for authentication (though not always successfully). And it is also a crucial component in realizing SkyNet and an Orwellian future (whichever you prefer). However, despite steady progress, the accuracy of face recognition by humans is so far unparalleled.
A study conducted by MIT researchers Pawan Sinha, Benjamin Balas, Yuri Ostrovsky, and Richard Russell looks at the ways people recognize each other in order to help guide computer vision research towards building better face recognizers. Their study “Face Recognition by Humans: Nineteen Results All Computer Vision Researchers Should Know About” identifies several characteristics that humans use to recognize faces.
Humans are able to handle significant degradations in images. Faces include Bill Clinton, Tom Hanks, and Jay Leno.
One of the results is that humans can recognize faces even in very low-resolution images. Eyebrows are one of the most important features for recognition. And our ability to tolerate degradations increases with familiarity. Read the full study here. The researcher’s website is found here.