FAMILIAR FACE RECOGNITION - A COMPARATIVE-STUDY OF A CONNECTIONIST MODEL AND HUMAN-PERFORMANCE

A connectionist simulation of familiar face recognition by humans, whi ch incorporates aspects of the Bruce and Young cognitive model, is dis cussed in detail. The computer model employs three self-organising neu ral maps, whose input vectors are the eye region, the mouth region and a reduced resolution full-face image of the input grey-scale face ima ges, together with a decision-making supervised multi-layer perceptron as the output network. A comparative set of experiments has been perf ormed using this simulation and human subjects for the task of familia r face recognition. When the gender of the facial images is supplied t o the simulation, the recall misclassifications of the simulation and the human subjects are statistically similar.