Mona Lisa probably wasn't a man, and it's even more unlikely that the famous artist who painted the famous painting used himself as the model.
So conclude University of Illinois researchers, who created a buzz last year when their facial-recognition software was used to analyze the Mona Lisa's enigmatic smile. (Conclusion: The model was happy, with touches of disgust, fear and anger, at least as Leonardo da Vinci painted her.)
Meanwhile, in the latest twist, UI Professor Thomas Huang has analyzed the painting to determine its subject's likely gender and compared it with a self-portrait of da Vinci.
The analysis, using the facial-recognition software developed by Huang and his students, says there's a 60-40 probability the painting is of a female. Even if it is a man, it doesn't match up well with Leonardo's sketch of himself.
"Of course, nothing proves anything," Huang, a UI electrical and computer engineering professor, said recently. "At least it indicates these conjectures may not be right."
The "conjectures" include the notion that da Vinci used himself or another man as the model for the painting.
The portrait, which da Vinci probably completed in 1507, is often said to be of Lisa Gherardini, wife of Francesco del Giocondo, a wealthy silk merchant from Florence, Leonardo's hometown. The painting's whole title is "Mona Lisa del Gioconda."
But various alternatives have been suggested with more or less justification through the years, including Leonardo himself, his mother, Mr. Giocondo, the Duchess of Milan Isabella of Aragon and a female friend of the powerful Florentine Giuliano de Medici.
A Bell Labs researcher used early photo morphing software to compare Mona with what's believed to be a da Vinci self-portrait and concluded that the artist was the model.
Huang, who co-leads the Human-Computer Intelligent Interaction group at the UI's Beckman Institute, is working on facial recognition for a variety of purposes.
Those include security systems and making your computer more personal by getting it to recognize you and set itself up accordingly, even adjust to the moods reflected on your face.
Businesses might use the system to collect customer data, like the ratio of guys to gals who order Quarter Pounders at McDonald's, or for "adaptive displays," billboards that change ads from ESPN to Victoria's Secret depending on whether a man or woman walks by, for instance.
Huang said the technology also might be used in "smart kiosks" – electronic bank tellers, for example – that could recognize and address a user properly.
In December, Nicu Sebe, a professor and sometimes collaborator of Huang's at the University of Amsterdam in the Netherlands, drew international attention by using software from the UI to examine the emotional state of the Mona Lisa.
Huang said the fun, as well as useful, test of the system has led to a small wave of requests to look at more paintings, showing that the technology may be useful to art historians and others in the humanities in addition to its other potential uses.
They've examined a self-portrait of the suicidal Vincent van Gogh, whom the software tagged as exhibiting profound sadness. They've also looked at Edvard Munch's "The Scream" for a Japanese TV station. The software pegged the dominant emotion in that painting's expressionist subject as surprise, not fear.
All this got Huang thinking about the historical speculation over the Mona Lisa's identity. He decided to apply the gender identification function of the system to the painting and also to compare it with da Vinci's self-portrait.
In the latter, Huang and his students used a database of facial photos to create a scale of the differences between two faces and then looked at how Mona and Leonardo's self-portrait fit on it. The two came out far more divergent than the average, indicating that the images don't appear to be the same person.
In the gender examination, they used the database to plot feature differences between males and females around the eyes, nose, mouth and elsewhere and then examined the Mona Lisa's face in the same way.
Mona's face, let's face it, is pretty matronly, so it's not surprising that the system made the odds 60-40 she was a woman. The result actually tends to validate the software.