Mona Lisa was not a man nor was she was a self-portrait by Leonardo Da Vinci. Exactly who and what she was may forever be shrouded in mystery, but researcher Thomas Huang of the Beckman Institute for Advanced Science and Technology is confident Da Vinci was not painting himself, or any other man for that matter.
Using gender recognition software he developed at Beckman, Huang said he and a group of students determined to a "maximum likelihood" that Mona Lisa was female and that her features are very dissimilar to those of Da Vinci.
Recently, a former colleague of Huang's in Amsterdam reported on the emotional component of Da Vinci's masterpiece using emotion recognition software developed largely at Beckman by Huang. Now Huang has applied gender recognition algorithms to an image of the Mona Lisa and an image of a sketch of Da Vinci to answer speculation that the subject was perhaps a male in disguise or that the artist was painting himself.
"Modern computer biometrics techniques have reached a point where face and related recognition problems can be solved with high accuracy under constrained conditions," Huang said.
Huang said he worked with students Zhenqiu Zhang, Jilin Tu, Shyam Rajaram, and Dennis Lin to carry out recognition experiments on the images. To do their comparisons, they used a face database that was ethnically diverse and split approximately 50-50 between males and females. Huang said texture can be taken into account in determining gender but was not in this case because algorithms used to account for cracks in the Mona Lisa image would also alter the facial image. So, only shape information was used for both experiments.
"Without getting into technicalities, let me just report some results of our computer experiments," Huang said. "The gender recognition algorithm indicates that the likelihood probabilities of Mona Lisa being female and male are respectively 60 percent and 40 percent. So a 'maximum likelihood' classifier will conclude that Mona Lisa is female."
By using distance measures of pairs of different faces in the database and comparing those standard and mean deviations with those of the Mona Lisa and Da Vinci, the researchers were also able to show that the distance measure between Mona Lisa and Da Vinci is much bigger than the average distance between the faces of two different persons, specifically by 2.5 standard deviations.
So, according to the computer, Huang said we could conclude that Mona Lisa is female and the faces of her and Da Vinci are not very similar.
"However, since there is no way to get the 'ground truth' we can keep speculating," he said.
The algorithms used in the study were developed with the support of an Advanced Research and Development Activity grant, the Yamaha Motor Company, and internal support from the Beckman Institute.