Science News for Students - Spring 2021

New device tells smiles from frowns — even through a mask It analyzes someone’s cheeks for telltale clues to expressions

capture that information from you. Such as from a camera. But it’s not convenient to always have a camera in front of your face. What if you’re exercising, cooking or shopping? Meanwhile, the sides—or contours —of someone’s cheeks change as they make different faces. These shapes tend to match specific expressions. Deep learning, an artificial-intelligence technique, can detect patterns like this. It just needs lots of practice, known as training. To train C-face, Chen and other team members made funny faces. Meanwhile, headphone cameras captured how the contours of their cheeks changed with each expression. And a camera in front of the face captured the locations of important “landmarks” around the eye- brows, eyes and mouth. The researchers then fed those landmark positions into a computer. It created a matching virtual version of the face that looked happy, sad, puzzled, surprised or something else entirely. Zhang’s group unveiled its new system in October 2020. The research- ers shared details about it at a virtual conference of the Association for Com- puting Machinery Symposium on User Interface Software and Technology. The pandemic shaped this research. Everyone’s face is unique. So each person who uses the C-face device has to train and test it on their own expres- sions. But in spring 2020, when they wanted to test this, the coronavirus pan- demic had sent most of the world into lockdown. So the team “got approval to conduct the studywith our room- mates,” says Benjamin Steeper, another Cornell student. The situation wasn’t ideal. Steeper converted one room of his apartment into “the science room,” complete with a desk, chair, cameras and everything else he needed. Chen set up his bedroom to also use for the study. He also stars in the video that the team used for training and test- ing C-face. The pandemic had another impor- tant impact on the team’s research. Face masks suddenly became part of

By Kathryn Hulick

Chen studies computer science in the lab of Cheng Zhang at Cornell University in Ithaca, N.Y. Zhang came up with the idea for this system. He calls it C-face. That “C” stands for facial contours. The C also is a pun because the device can “see” your face. His team’s goal is to create technol- ogy that can better understand people. Right now, our devices are mostly clue- less about howwe’re feeling or what we need. But over time, more devices will understand us. Zhang hopes that “ev- erything will be smart in the future.” For example, your phone might recognize when your face looks upset and suggest calming music. For your phone to know you’re upset, however, it has to somehow

Tuochao Chen gapes. Then he sneers. Then he grimaces. As he makes faces, he wears a device that looks like a pair of headphones. But instead of play- ing sound, it points cameras toward his cheeks. The cameras only see the sides of his face. Surprisingly, that’s enough facial real estate to tell a sneer from a smile, or a laugh from a frown. A computer system connected to the headphones can figure out what Chen’s eyes and mouth look like without seeing them directly. It works even if Chen wears a face mask. The system can tell if he’s smiling or frowning behind it.

When you wear a mask, other people often can’t tell if you’re smil- ing or frowning. But a new device called C-face can. It learns our facial expressions by watching the mo- tion of our cheeks.

LIFESTYLEVISUALS/E+/GETTY IMAGES PLUS

34 SCIENCE NEWS FOR STUDENTS | Invention & Innovation