Linear models of visual search are highly implausible: towards a better understanding of search in real world scenes using logarithmic search functions.

摘要

Lleras, Cronin & Buetti (submitted) proposed an Information Theory of Vision (ITV) that describes visual search as a combination of two sequential stages: attentional screening (driven by dissimilarity and logarithmic in nature) and attentional scrutiny (mediated by working memory and linear in nature). ITV is meant to capture both search in traditional experiments as well as search in real world scenes. A crucial prediction of this theory is that, based on information theory (Shannon, 1947) and Hick's law, the duration of the screening stage should be approximately logarithmic in terms of total setsize because processing time is proposed to be proportional to the amount of information in a display. Further, ITV proposes that only a subset of elements in the scene (candidates) produce a linear processing cost. An approximation of the reaction time formula (for large set sizes) is: RT=a+D*ln(setsize)+I*Nc Here, we compared the plausibility of ITV to that of theories of visual search that propose RT is a linear function of the number of items in the display (or a subset of them). We borrowed data from Wolfe et al. (2011) (data from Experiment 2), and performed a parameter estimation analysis comparing our model with traditional linear model: RT=a+I*Nc. In current theories of visual search, both inspection time I and the number of candidates Nc are thought to vary with each search scene, while in our model we fixed I based on data from Wolfe et al. (Experiment 3). Thus, both models have an equal number of undetermined parameters. We computed parameter pairs that gave minimum squared residuals with equal constant a and computed plausibility by finding the proportion of estimated parameters that fall within empirically observed ranges. Our results show that linear models of search are highly implausible whereas our model is highly plausible.