Recommendations are independently chosen by Reviewed’s editors. Purchases you make through our links may earn us a commission.
March 23, 2007 – Ever wonder what Brad and Tom's love child would look like? Ashton and Demi’s future kids? McDreamy and yourself? A new website called ModiFace, launched by researchers at the University of Toronto, allows users to fuse their portrait shots with the faces of their favorite celebrities or with other home photos, based on a face detection algorithm. With a database of celebrity photos, ModiFace is geared for those curious about pairing up tabloid stars or those simply looking for a new hairstyle.
Since the site launched in December 2006, Modiface has received 100,000 hits per day with its peak at 2 million visitors in 24 hours and 5 million users in one week.
The ModiFace website grew from a larger project. Initially developed for cosmetic surgery practices, researchers at the University of Toronto have spent the last two years in development. ModiFace evolved into an online-based community looking to have fun with photos.
Existing cosmetic surgery "software suffered from one issue; it had to be manually utilized," said University of Toronto Professor Parham Aarabi of the Artificial Perception Laboratory in an interview with DigitalCameraInfo.com. The researchers automated imaging for face modification using face detection technology.
Digital camera users are already familiar with the term "face detection," as it has been implemented by manufacturers including Canon, Pentax, Olympus Fuji, Samsung, and more recently Sony and GE.
"Most digital camera face detection algorithms are very similar [to ModiFace]," said Aarabi. ModiFace uses a generic face template that most front-facing human faces fit into.ModiFace uses a series of steps including face detection that first evaluates the nose and then* feature* detection that evaluates the eyes and lips, according to the abstract "Face Fusion: An Automatic Method for Virtual Plastic Surgery." After finding the three imperative original features highlighted by rectangles, those features are resized and replaced into the model face, based on probability. The shifting and blending steps follow to make it look the model face appear natural. As far as color goes, ModiFace uses RGB values and can, for the most part, account for all skin types. While ModiFace’s face detection process has a 98 percent accuracy rate, the color adjustment rate of accuracy is only 60 percent. For example, to combine the face shape of Oprah with the eyes of Paris Hilton, there is a 40 percent chance that the Caucasian skin color around the eyes will not match the African-American skin tone of the rest of the face.
When asked if ModiFace could work with animals, Aarabi said that it could work selectively. "You might be able to put Britney Spears’s eyes on your pet’s face," said Aarabi.
Future plans for ModiFace include 3D processing. Currently, ModiFace only works with frontal portraits but could soon account for side and angled face shots, said Aarabi.
ModiFace is funded by Canadian governmental grants, although one day, the project may be commercialized. Users need not worry about losing free access to ModiFace. Online users and plastic surgeons will most likely still be able to experiment, although may notice a few ads on the site, but not for a while, according to Aarabi.
Interested readers can find the full abstract at http://www.apl.utoronto.ca/publication/i/rabi_fusion06.pdf. Users can test out ModiFace for themselves at http://www.modiface.com/index.php.