DescriptionThis digital collage is made up of over 7,000 faces, all automatically extracted from a selection of Bertillon cards held by the Ohio History Connection. The images are arranged in order of facial similarity by the OpenFace facial recognition package. OpenFace uses an artificial neural network to map images onto a 128-dimensional hypersphere—that is to say, it reduces each image to a series of 128 numbers that then represent its overall appearance. These numbers can then be used for comparison between the images and to differentiate between them. A close look at this grid reveals clusters of faces that do indeed appear to be similar to the human eye, others are less strikingly so. The measures of similarity produced by OpenFace’s neural network do not always necessarily correspond to our everyday lived experience of visual likeness.
Researchers: Sam Nosenzo, Alison Langmead and Aaron Henderson