FAITHFUL REPRESENTATION OF SIMILARITIES AMONG 3-DIMENSIONAL SHAPES INHUMAN VISION

Efficient and reliable classification of visual stimuli requires that their representations reside in a low-dimensional and, therefore, comp utationally manageable feature space. We investigated the ability of t he human visual system to derive such representations from the sensory input-a highly nontrivial task, given the million or so dimensions of the visual signal at its entry point to the cortex. In a series of ex periments, subjects were presented with sets of parametrically defined shapes; the points in the common high-dimensional parameter space cor responding to the individual shapes formed regular planar (two-dimensi onal) patterns such as a triangle, a square, etc. We then used multidi mensional scaling to arrange the shapes in planar configurations, dict ated by their experimentally determined perceived similarities. The re sulting configurations closely resembled the original arrangements of the stimuli in the parameter space. This achievement of the human visu al system was replicated by a computational model derived from a theor y of object representation in the brain, according to which similariti es between objects, and not the geometry of each object, need to be fa ithfully represented [Edelman, S. (1995) Minds Machines 5, 45-68; cf. Shepard, R. N. (1968) Am. J. Psychol. 81, 285-289].