Abstract
To date only a small palette of isotrigon textures have been available to study how the brain uses higher-order spatial correlation information. We introduce several hundred new isotrigon textures. Previous isotrigon textures were binary (black and white) but the new versions have 3 or more levels or colours. Some of these textures have properties that allow simple means for temporal modulation of their higher order statistical properties. These modulation properties can be used in evoked potential or fMRI studies to extract neural responses to higher order spatial correlations. We also examine the issues of how many textures make an adequate training set, and how representative individual examples are of their texture class, depending upon the size of the examples. This is done through examining the convergence rates of the mean second and third order correlation functions, and by consideration of the so called mini-texture spectrum. Human discrimination of 90 of these patterns from uniformly distributed ternary noise was also quantified in a series of 2AFC trials. Discrimination was examined for 3 texture sizes: 8, 16 and 32 pixels squared. When presented the textures were surrounded by uniformly distributed ternary noise. The stimuli and background were presented for 200 ms and 16 repeats were obtained at each size. Simple models were constructed that can closely mimic human discrimination performance providing at least a fourth-order classifier is used. The modelling study indicates however that more than one discrimination mechanism is required, to capture human performance over the textures examined.
This research was supported by the ANU Centre for Visual Sciences and the Australian Research Council through the ARC Centre of Excellence in Vision Science (CE0561903)