Our Grins Have Grown Wider Over the Last Century, Machine-Vision Algorithms Show
In the past we said ‘prunes’ not ‘cheese’ when we smiled for the camera.
Image: University of California, Berkeley.
When looking through old school yearbooks, we usually notice outdated hairstyles and clothes, but a new arxiv study featured in the MIT Technology Review, reveals that people's smiles also change over time.
But now, researchers headed up by Shiry Ginosar, a computer scientist at the University of California, Berkeley, have created machine-vision algorithms that can see changes in clothes and hairstyles over the decades. It can also detect how our smiles have become more intense over the years.
For the study, the team unleashed their machine-vision algorithms on over a century of digitized US high school yearbook photographs. They used 37,921 images—which showed subjects in a front-facing pose—from more than 800 yearbooks. The researchers divided the portraits by decade and merged the photographs to create an "average" face, which helped them work out the standard facial expressions, hairstyles and clothing for each decade.
The researchers used their algorithms to measure changes in lip curvature, finding a trend that, over the last century, our photo grins have become more intense. While people wore the same neutral expression that they would have used for a painted portrait in the years following photography's invention, this changed with the rise in the medium's popularity, and the advent of advertising, which featured models with less neutral expressions.
"Etiquette and beauty standards dictated that the mouth be kept small—resulting in an instruction to "say prunes" (rather than cheese) when a photograph was being taken," write the researchers in their paper. "All of this changed in the 20th century when amateur photography became more widespread."
The machine-vision algorithm also corroborated previous findings on smiles in photography—for example that women grin more frequently than men. Previous research on this was conducted manually. "By use of a large historical data collection and a simple smile-detector we arrived at the same conclusion with a minimal amount of annotation and virtually no manual effort," the researchers write.
While the researchers were able to glean insights on the evolution of smiles, hairstyles and clothes they recognize their dataset is limited. For instance, the African-American population are absent from the photographs until the middle of the 20th century.
Next up, the researchers want to make their algorithms more specific, allowing them to measure things like how long certain fashion fads lasted.
- University of California
- data mining
- motherboard show
- machine vision
- Shiry Ginosar
- say cheese
- say prunes