In this report we discuss a variety of psychophysical experiments that
explore different aspects of the problem of object recognition and re
presentation in human vision. In all experiments, subjects were presen
ted with realistically rendered images of computer-generated 3D object
s, with tight control over stimulus shape, surface properties, illumin
ation, and viewpoint, as well as subjects' prior exposure to the stimu
lus objects. Contrary to the predictions of the paradigmatic theory of
recognition, which holds that object representations are viewpoint in
variant, performance in all experiments was consistently viewpoint dep
endent, was only partially aided by binocular stereo and other depth i
nformation, was specific to viewpoints that were familiar, and was sys
tematically disrupted by rotation in depth more than by deforming the
2D images of the stimuli. The emerging concept of multiple-views repre
sentation supported by these results is consistent with recently advan
ced computational theories of recognition based on view interpolation.
Moreover, in several simulated experiments employing the same stimuli
used in experiments with human subjects, models based on multiple-vie
ws representations replicated many of the psychophysical results conce
rning the observed pattern of human performance.