Assessing Algorithms as Computational Models for Human Face Recognition

We assessed the qualitative accord between several automatic face recognition algorithms and human perceivers. By comparing modeland human-generated measures of the similarity between pairs of faces, we were able to evaluate the suitability of the algorithms as models of human face recognition. Multidimensional scaling (MDS) was used to create a spatial representation of the subject response patterns. Next, the model response patterns were projected into the MDS space. The results revealed a bimodal structure for both the subjects and for most of the models, indicating that the qualitative performance of the subjects and models was related. The bimodal subject structure reflected strategy differences in making similarity decisions. For the models, the bimodal structure was related to combined aspects of the representations and the distance metrics used in the implementations.