A number of parametric texture models operate by assuming a plausible image representation for the early primate visual system, decomposing the target image into some number of frequency and orientation bands (Cano & Minh,
1988; Heeger & Bergen,
1995; Malik & Perona,
1990; Porat & Zeevi,
1989; Portilla & Simoncelli,
2000; Simoncelli & Portilla,
1998; Zhu et al.,
1998). The spatially averaged responses in some combination of these bands form the parameters of the model, whose values are then matched by the synthesis procedure. The parametric texture model of Portilla and Simoncelli (Portilla & Simoncelli,
2000; Simoncelli & Portilla,
1998) extended this approach by additionally matching the correlations between channels and other statistics, producing more realistic appearance matches to textures. This model has since had broad impact on the field of human perception and neuroscience: the texture statistic representation may provide a fruitful way to understand the processing in mid-ventral visual areas (Freeman & Simoncelli,
2011; Freeman, Ziemba, Heeger, Simoncelli, & Movshon,
2013; Movshon & Simoncelli,
2014; Okazawa, Tajima, & Komatsu,
2015; Ziemba, Freeman, Movshon, & Simoncelli,
2016), and it has been argued to provide a good approximation of the type of information encoded in the periphery, and thus a model for tasks such as crowding and visual search (Balas, Nakano, & Rosenholtz,
2009; Freeman & Simoncelli,
2011; Keshvari & Rosenholtz,
2016; Rosenholtz,
2011; Rosenholtz, Huang, & Ehinger,
2012; Rosenholtz, Huang, Raj, Balas, & Ilie,
2012)—though other evidence questions the more general adequacy of this representation for explaining crowding and peripheral appearance (Agaoglu & Chung,
2016; Clarke, Herzog, & Francis,
2014; Herzog, Sayim, Chicherov, & Manassi,
2015; Wallis, Bethge, & Wichmann,
2016).