Teaching computers how to interpret visual data is going to be essential to the development of things like self-driving cars and mood-sensing technologies? Humans assess the world around us mainly through vision, and computers are starting to do the same. Digital artist Adam Ferriss works with facial recognition software and pushes it to the extreme, creating portraits that reveal the strange beauty in a machine’s search for meaning.
For his project, Ferriss took a facial recognition algorithm called SURF, which is used to identify “interesting” parts of an image, also known as feature detection. By turning up the settings so that the feature detection is way more sensitive than necessary, the portraits become a mere backdrop for thousands of lines and circles indicating features where none are apparent.
“One of my friends, Jesse Fleming, mentioned that it was interesting that in the algorithm’s overzealous search for features, it obscures the thing it was initially searching for,” Ferriss says.
It’s true: with so many lines and circles crowding these images, seeing the face underneath becomes almost impossible. This highlights how differently computers process information. When we see a face, our brain recognizes it by looking at the pattern of features in conjunction with each other. We don’t consciously waste time looking at the eyebrow hairs of a friend before determining who they are. But a face-shaped image means nothing to a computer without meticulously interpreting every data point. Perhaps, with more advancement in this kind of technology, computers will one day intuit facial features as easily as we do.
“The circles here represent individual features, with their size indicating the confidence the computer has in the feature, and lines pointing in the angle the algorithm believes the feature is pointing,” Ferriss says.
Ferriss’ project isn’t a practical demonstration, but a provocation, showing us how these increasingly prevalent programs function. “This kind of over-detection would be pretty useless in any real application,” he says. “I’m interested in the visual language of identification that computer scientists have created to express what their programs are doing.” In other words, his portraits let us to see through a computer’s eyes.