Visual features as carriers of abstract quantitative information
Ronald A. Rensink
University of British Columbia
Journal of Experimental Psychology: General, in press.
Four
experiments investigated the extent to which abstract quantitative information can
be conveyed by basic visual features. This
was done by asking observers to estimate and discriminate Pearson correlation
in graphical representations where the first data dimension of each element was
encoded by its horizontal position, and the second by the value of one of its
visual features; perceiving correlation then requires combining the information
in the two encodings via a common abstract representation. Four visual features were examined: luminance,
color, orientation, and size. All were
able to support the perception of correlation.
Indeed, despite the strikingly different appearances of the associated stimuli,
all gave rise to performance that was much the same: just noticeable difference
was a linear function of distance from complete correlation, and estimated
correlation a logarithmic function of this distance. Performance differed only in regards to the
level of noise in the feature, with these values compatible with estimates of channel
capacity encountered in classic experiments on absolute perceptual magnitudes. These results suggest that quantitative information
can be conveyed by visual features that are abstracted at relatively low levels
of visual processing, with little representation of the original sensory
property. It is proposed that this is achieved
via an abstract parameter space in which the values in each perceptual
dimension are normalized to have the same means and variances, with perceived correlation
based on the shape of the joint probability density function of the resultant elements.