We take it for granted that, when someone takes our photo, we should smile for the camera. But if Instagram had existed a century ago, it's highly likely that stone-faced selfies rather than the beaming faces would dominate our photo feeds.
The smiles we turn on for the camera today are roughly 10 times bigger than the smiles photographed 100 years ago, according to new research from a team of computer scientists and software engineers at the University of California, Berkeley and Brown University in Rhode Island.
The photo montage, above, from their paper A Century of Portraits: A Visual Historical Record of American High School Yearbooks, shows how our smiles grew over the 20th century.
It was generated by using computer algorithms to analyse nearly 38,000 American high school yearbook photos taken between 1905 and 2013. The photos were grouped by year and superimposed to create an "average" male and female face for each decade.
The chart below shows the same information, using lines instead of pictures. It shows a rapid increase in the intensity of smiles between the 1900s and 1950s, a trend that continues today.
It also shows women have consistently smiled more than men. In fact, men began the century with a -0.5 degree lip curvature, otherwise known as a slight frown.
Why our frowns were gradually turned upside down is closely tied to the invention of the camera.
In the late 19th century, people posing for photographs adopted the same habits as people posing for painted portraits. For paintings, a serious expression made sense because it was easier to maintain over an extended period of time.
Also, small mouths were considered more beautiful and dignified, which was why people were instructed to say 'prunes' (rather than 'cheese') when their photo was being taken, according to the paper.
But all this changed, thanks to the spread of amateur photography and a century-long Kodak advertising campaign that promoted the idea that photography was for happy occasions, such as holidays.
From "prunes" (left) to "cheese" (right): Lip curvature was used to measure the intensity of smiles. Photo: Ginosar et al. 2015
This history of the smile has been explored in previous research, which used manual methods to analysis sample images. The team from the University of California and Brown University set out to verify these claims using a far larger data set and machine-vision algorithms.
Behold, above left, the tool the team devised to measure the "smile intensity" of each photo.
This "lip curvature metric" measures the angles at the corners of the mouth - basically, how far the corners of the mouth curve upwards.
The next two images below show the men and women with the smile closest to the average smile of the 10-year period, from 1905 to 1955 and then 1965 to 2005. Notice how our smiles grow as the decades pass.
1905 to 1955: The photos with the closest smile to the average for that 10-year period. Photo: Ginosar et al. 2015
The algorithm also picked up the changing hairstyles of the 20th century, documenting "[t]he finger waves of the '30s. The pin curls of the '40s and '50s. The bob, 'winged' flip, bubble cut and signature glasses of the '60s. The long hair, Afros and bouffants of the '70s. The perms and bangs of the '80s and '90s and the straight long hair fashionable in the 2000s", according to the paper.
"The most interesting part for me was that we found such a strong correspondence between the results we get from machine learning algorithms in mining a large data set, and the results of painstaking manual work in history and sociology," lead author Shiry Ginosar told Fairfax Media.
The features of different time periods that prompt people to say, "This person is from the '60s," are often clear to us visually but hard to capture in words, she said.
1955 to 2005: The photos with the closest smile to the average for that 10-year period. Photo: Ginosar et al. 2015
"Yearbooks are a cool set of data as they give us kind of a 'peephole' into the past where everything - pose, the reason for taking the picture, the age of the subjects etc. - remains constant, except for the passing time."
This allows researchers to "tease out" the visual cues that change, such as smiles or hairstyles, Ms Ginosar said.
One limitation of the research, which the authors highlight, is that African Americans were poorly represented in high schools until the middle of the 20th century, creating a distinct bias in the data set.
Nevertheless, the findings are powerful. The rapid spread of photography over the past 200 years has created an enormous data set that until now, has been mostly untapped.
Data mining means we can finally access the information locked in these massive visual data sets - and without the painstakingly manual effort of the past.
Now that's something to smile about.
