Machine identification of human faces

摘要

Improved techniques for automatic recognition of human face profiles are reported. A computer file of 112 subjects was constructed, each face description being a 17-dimensional feature vector whose components were formed from the averages of three sittings. When a fourth pose was introduced as an ‘unknown’ for file identification, recognition accuracy was 96%. In the 4% of failures, the correct file entry was in second place in the rank-ordered list of Euclidean distances. For experiments of asserted identity, where the unknown claimed to be a particular file member, confirmation was essentially perfect, as was rejection of non-file interlopers.