Abstract
Abstract: :
Purpose: Certain kinds of local image statistics are processed rapidly and in parallel. Two such statistical cues, local first-order statistics and certain kinds of local fourth-order statistics, are also represented in visual working memory. Here we investigate whether these statistical cues are represented in a continuous fashion, or a categorical one. Methods: We constructed examples of 8x8 arrays of black and white checks that typified a range of statistical structure along a continuum (parameterized by a quantity c). For first-order statistics, the continuum was from all-black (c=-1) to all-white (c=+1). For fourth-order statistics, the continuum was from "odd" (c=-1) to "even" (c=+1). Stimuli consisted of four of these arrays (S1, 600 ms), followed by a blank interval (200 ms), followed by S2 (200 ms). Subjects were asked (4-AFC without feedback) to identify the array in S2 in which 16 of its 64 checks changed (the "target"). In each trial, arrays were independently created to typify either of two neighboring values of c separated by a small amount (0.25 for first-order statistics and 0.4 for fourth-order statistics). In half of the trials, the change in the target represented a small change in the statistics parameterized by c. In the other half of the trials, the statistics parameterized by c was unchanged between S1 and S2. Results: For first-order statistics, subjects' (N=4) fraction correct was approximately 70% when there was no change in statistics, and 80% when there was a change. The improvement associated with a change in statistics was uniform across the range of c tested, as was the average reduction in reaction time (ca. 50 msec). For fourth-order statistics, subjects' (N=6) fraction correct was approximately 65% across most of the range of c, independent of whether there was a change in statistics between S1 and S2. The improvement associated with a change in statistics was restricted to the extreme "even" end of the continuum, and there was no consistent reduction in reaction time. Conclusion: These observations are most consistent with a continuous, rather than categorical, representation of image statistics in visual working memory.
Keywords: 626 visual search • 597 texture • 466 memory