Abstract
Crowding refers to phenomena in which a task such as object recognition becomes difficult in the periphery due to the presence of flanking stimuli. We may be unable to identify a letter when it is flanked by other letters, or unable to identify a face seen peripherally. Subjectively, we perceive “jumble” of features. Researchers suggest that crowding is due to feature integration over an inappropriately large region (Pelli et al, 2004), or to “compulsory texture perception” (Parkes et al, 2001). For displays consisting of simple stimuli such as oriented Gabors, such texture perception may amount to perception of orientation statistics such as mean orientation (Parkes et al, 2001). But what does compulsory texture perception mean for arrays of letters, for faces, or for natural scenes viewed peripherally?
Many texture synthesis algorithms generate new samples of a texture from a set of image statistics taken from a sample texture patch. Texture synthesis makes an explicit statement about what “texture processing” means for the visual system: If texture synthesis adequately generates new sample textures, the statistical properties it employs are sufficient to describe the texture percept. Recent work, particularly (Portilla & Simoncelli, 2000, inspired in part by Heeger & Bergen, 1995) constrains synthesized textures by statistics of biologically plausible features. The generated texture is often difficult to distinguish from the original texture, particularly when viewed briefly or somewhat peripherally (Balas, 2006).
Synthesized textures may be generated from arbitrary stimuli, to visualize what “compulsory texture perception” of these stimuli might mean. These “texturized” images subjectively capture some of the percept of the original stimuli viewed under crowded conditions. Furthermore, we demonstrate that difficulty performing a task, such as letter recognition, in foveated texturized displays is correlated with difficulty in performing that task with the original displays under crowding conditions.