Contributed Talks III: Surface color information gained in searching natural scenes

Natural scenes are often spatially and spectrally complex, containing fields, woodland, grasses, ferns, and flowers, as well as buildings and treated surfaces. How we search such scenes may not be readily explained by notions of object salience. Instead, gaze behavior may satisfy a simpler requirement: to gain information about a scene's distinct reflecting elements, characterized by their reflected color. This idea was tested by estimating the information from seven observers' eye fixations recorded while searching natural outdoor scenes for a small neutral target. Scenes were rendered on a computer-controlled display and appeared repeatedly in successive trials of 1 s each. Information was estimated numerically between reflected spectral radiances at the pooled fixation positions and the corresponding excitations in long-, medium-, and short-wavelength-sensitive cone photoreceptors. These estimates were then compared with those from the same number of sample points distributed randomly across each scene. In 16 scenes out of the 20 tested, fixations delivered more information than random sampling, the proportion depending little on photoreceptor noise. The information gain in each scene covaried with color entropy, a frequency-weighted measure of spectral diversity. Although information gain may not account fully for search behavior, it could offer a basis, where appropriate, for more object-oriented explanations.