Marsha Meytlis and Lawrence Sirovich On the Dimensionality of Face Space, Transactions on Pattern Analysis and Machine Intelligence (2007), v. 29(7), pp. 1262-1267 (featured article of the Month) [Cited by 15.]
The dimensionality of face space is measured objectively in a psychophysical study.
Within this framework we obtain a measurement of the dimension for the human visual system.
Using an eigenface basis, evidence is presented that talented human observers are able to
identify familiar faces that lie in a space of roughly 100 dimensions, and the average
observer requires a space of between 100 and 200 dimensions. This is below most current
estimates. It is further argued that these estimates give an upper bound for face space
dimension, and this might be lowered by better constructed "eigenfaces", and by talented
Lawrence Sirovich and Marsha Meytlis (2009) Symmetry, probability and recognition in face space, PNAS (2009), v.106(17), pp.6895-6899 [Cited by 3.]
The essential midline symmetry of human faces is shown to play a key role in facial coding and recognition. This also has deep and important connections with recent explorations of the organization of primate cortex, as well as human psychophysical experiments. Evidence is presented that the dimension of face recognition space for human faces is dramatically lower than previous estimates. One result of the present development is the construction of a probability distribution in face space that produces an interesting and realistic range of (synthetic) faces. Another is a recognition algorithm that by reasonable criteria is nearly 100% accurate.
Marsha Meytlis, A Model of Internal Face Space, (accepted by Visual Cognition)
Faces vary in their distinctiveness, and it has been suggested that more distinctive faces are
located further from the mean face in internal or perceptual face space (Valentine, 1991), than
typical faces. This paper proposes a pixel-based computational model of face distinctiveness. I
show that the notion of psychological distinctiveness is correlated with the metrical Euclidean
distance, from the mean face. Distances are measured in in pixel-based face space, which serves
as the basis of the computational model. A prediction of the model is that pair-wise perceptual
distances between distinctive faces are, on average, greater than between typical faces. This is
confirmed by psychophysical experiment.
*Marsha Meytlis, *Zachary Nichols, and Sheila Nirenberg Correlations play a negligible role in coding white noise and natural scene stimuli in complete retinal populations, (submitted to Journal of Neuroscience) (*contributed equally to the work)
The role of correlated firing for representing information has been a subject of much discussion. Several studies in areas including retina, visual cortex, somatosensory cortex, and motor cortex, have suggested that it plays only a minor role, contributing less than 10% of the total information carried by the neurons (Gawne and Richmond, 1993; Nirenberg et al., 2001; Oram et al., 2001; Petersen et al., 2001; Rolls et al., 2003). A limiting factor of these studies, however, is that they were carried out using pairs of neurons; how the results extend to large populations is not clear. Recently, new methods for modeling network firing patterns have been developed that open the door to addressing the importance of correlating firing in answering this question for more complete neuronal populations (Truccolo 2005; Pillow et al. 2008). Pillow et al. (2008) used this approach to assess the importance of noise correlations in primate retina, using random checkerboards as the stimuli. His results showed that the inclusion of such correlations produced only a modest (~20%) increase in the amount of information, an increase considerably smaller than what might be expected from extrapolations using the pairwise data. Here we show that this finding generalizes in two ways. First, it generalizes across species: in mouse ganglion cell responses to random checkerboards, correlations were also found to increase information by a small amount (~14%). More importantly, it applies to natural movies, and in a stronger form: for these stimuli, correlations were found to contribute almost no additional information.
Marsha Meytlis and Cheuk Tang Lie Detectors for Face Recognition, (in preparation)
Meytlis, M., Bomash, I., Pillow, J., and Nirenberg, S. Assesssing the importance of correlated firing using large populations of neurons, Society for Neuroscience (SFN) poster (2009)