Abstract

Human texture discrimination seems to be limited to measurements of local differences across receptive field (RF) activities. Important RF properties include spatial frequency and orientation selectivity, where each RF generates a filtered map of the input image. Significant local differences within each map can be taken as evidence for texture boundaries. A model that combines odd and even symmetry RF maps, generating energy maps, produces discrimination values that correlate well with human performance.¹ Here we consider the role of spatial noise (internal and external) in filtered maps as a performance limiting factor. We show that external noise, mainly due to randomization of texture parameters (like orientation), creates spurious local differences in the filtered maps (due to their orientation selectivity). This noise level depends on the particular texture element (e.g., its rotational symmetry) and may have a stronger distractive effect when it is large in the background. When quantified, this improved model accounts for asymmetry in human texture discrimination performance.²

© 1989 Optical Society of America