Communications of the ACM
Neural Network Learns to Identify Criminals By Their Faces
Credit: Pinterest
Researchers from China's Shanghai Jiao Tong University applied a variety of machine-vision algorithms to study faces of criminals and noncriminals. They took still photos of 1,856 Chinese men aged 18 to 55 with no facial hair, half of whom were criminals. Investigators then used 90 percent of these images to train a convolutional neural network to recognize the difference and tested the neural network on the remaining 10 percent of the images. The neural network was able to correctly identify criminals and non criminals with an accuracy of nearly 90 percent.
"These highly consistent results are evidences for the validity of automated face-induced inference on criminality, despite the historical controversy surrounding the topic," according to the researchers. The neural network uses three distinct facial features to make its classification: the curvature of the upper lip, which is on average 23 percent larger for criminals than for noncriminals; the distance between the inner corners of the eyes, which is 6 percent shorter in criminals; and the angle between two lines drawn from the tip of the nose to the corners of the mouth, which is 20 percent smaller in criminals.
The researchers also found that the data for criminal faces has much greater variance than the data for noncriminals. "In other words, the faces of general law-biding public have a greater degree of resemblance compared with the faces of criminals, or criminals have a higher degree of dissimilarity in facial appearance than normal people," the researchers conclude.
From Technology Review
View Full Article
Abstracts Copyright © 2016 Information Inc., Bethesda, Maryland, USA
No entries found