Purchase this article with an account.
Igor Utochkin; Ensemble summary statistics as a basis for visual categorization. Journal of Vision 2015;15(12):891. doi: 10.1167/15.12.891.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
A core idea beyond ensemble perception is that the visual system is capable of representing features of multiple objects as abstract statistical entities – distributions characterized by the mean, deviation, numerosity, etc (Alvarez, 2011). This approach gives insights into some more complex processes, such as categorization – perceiving some items as representing the same or different types of objects. Even in highly heterogeneous displays (e.g., cars in traffic or berries among leaves), ensemble summary statistics can provide massive categorization, probably via distribution comparisons which are akin to statistical tests. In first-order categorization, the visual system decides whether multiple features along a single dimension represent same or different categories. A plausible way of doing that is testing the distribution for normality. If all pooled features can be collected under the same Gaussian, then they probably represent the same category; but if the test fails showing normality then the distribution is decomposed into several Gaussians, each representing a separate category. Behavioral and neural data show that the results of normality test may depend on feature separation in an ensemble (Treue et al., 2000; Yurevich & Utochkin, 2014). In second-order categorization, the visual system operates over separate distributions (as the result of the first-order processing). Second-order statistics can be analyzed in-depth, when one of first-order distributions (e.g., by color) is attended and a new dimension (e.g., size) is analyzed within it. On the other hand, second-order processing can be in-extent, when few selected distributions are compared with each other (say, if blue objects are larger on average than yellow). The latter type of processing behaves like standard statistical tests, such as compare means or proportion comparisons (Fouriezos et al., 2008; Utochkin, 2013). Anyway, it appears that second-order processing is severely limited by attentional and working-memory capacities (Attarha et al., 2014; Halberda et al., 2006).
Meeting abstract presented at VSS 2015
This PDF is available to Subscribers Only