Perception of natural images with exponential luminance frequency distributions

Abstract

The luminance frequency distribution of many rural scenes can be approximated by an exponential function (Pinto, 2004, MPhil thesis, UMIST, Manchester , UK) but, in psychophysical experiments where luminance frequency distributions were manipulated, observers seem to prefer an exactly exponential dependence over the original approximately exponential dependence (Pinto et al, 2005 Investigative Ophthalmology & Visual Science 46 E-­‐Abstract 4574). The purpose of this work was to investigate what properties of the scenes determine such preference. Images of rural scenes were obtained with a hyperspectral imaging system (Foster et al, 2004 Visual Neuroscience 21 331 -­‐ 336) and, for each image, its luminance frequency distribution was computed and a set of images with exactly exponential distributions but with different coefficients was derived. The images resulting from these manipulations were displayed on a calibrated colour monitor. In each experimental trial, the observer had to express a preference between the original image and a version with an exactly exponential luminance frequency distribution. The pattern of responses was analysed and correlated with the luminance properties of the scenes. It was found that the degree of preference for an exact exponential function was stronger for distributions with intermediate widths and weaker for short -­‐ and long-­‐tailed native luminance frequency distributions. These data suggest that the visual system may be optimised for exponential luminance distributions with specific decay parameters