Doppelgänger app – Can someone unlock your iPhone?
Could your doppelgänger trick your iPhone’s facial recognition feature into believing that you are the same person? The answer might lie within our newly-built facial recognition software "Doppelgänger app" at Ailabs.tw.
One of social media's hottest topics is "How can two celebrities, without any blood relation, look identical?" This discussion went viral on PTT, one of Taiwanese largest bulletin board system (BBS), right after Apple released the "Face ID" feature with iPhone X in November, 2017. Many people were wondering: Can Elva Hsiao(蕭亞軒) unlock Landy Wen(溫嵐)'s iPhone?
Another hilarious post on PTT that collects all kinds of collage images made from "LookAlike Celebrities" raised AILabs.tw's curiosity - If human eyes cannot tell the difference between two faces, can artificial intelligence do a better job at telling them apart than human eyes? After some quick internal discussions, we kickstarted the "Doppelgänger app" project.
The app compares two faces and measures facial similarity from the images uploaded by users. It is designed to be a simple, fast and fun app. Within a few steps, you are able to find out how much you look like your friends, family members or even your favorite celebrities.
The UI flow of the App:
Once the photos are uploaded, the app shows up a score that represents the degree of similarity between the two faces. This score is calculated by our AI algorithm and ranges from 0 - 100. The higher the score, the greater the similarity.
After receiving the result, the app prompts a input message box for users to give their comments about the result. This is usually the funniest part when you realize how similar you are with one of your childhood friends or how the app mistakes you for an all-time rock star.
The last step is to share the result on your social network or via instant messenger to tell your friends about your new discovery.
When it comes down to measuring the granular level of pieces of your face, the app observes subtle differences between two faces, such as the distance between your nose, mouth and the contour of your ear.
Pieces of these facial features, represented by a 128-dimensional vector, are grouped together by a computer's algorithm to form higher level objects, such as the eyes or the chin. Individual objects are put together to become a higher level concept, such as the face. At each level of object representations, detailed facial features are examined and compared, allowing the app to distinguish one person from another.
Behind the Doppelgänger app is our facial recognition system which applies deep learning to empower the machine's capability. For more details about how our system works, please visit this blog post.
With the spirit of "Myth-buster", we took some examples from Taiwan, England and the U.S. to see if AI can tell the differences among human faces. We use the photos of celebrities who look alike and are often mistaken for someone else.
Elva Hsiao(蕭亞軒) v.s. Landy Wen(溫嵐)
Natalie Portman v.s. Keira Knightley
Those considered to be “twin celebrities” usually get scores around 50. From the machine's perspective, these two people do not look remotely similar. It turns out that it is not easy to trick our apps, after all.
Contribution to the open-source community
We have released Android and iOS app. We also open-sourced this app's source code and API for similarity score calculation on Github. Check it out!
Leave a ReplyWant to join the discussion?
Feel free to contribute!