Figure_4.tif (297.54 kB)

Histograms of total luminance in target and distractor images as a function of the number of iterations.

Download (0 kB)
posted on 20.02.2013, 14:00 by Vadas Gintautas, Michael I. Ham, Benjamin Kunsberg, Shawn Barr, Steven P. Brumby, Craig Rasmussen, John S. George, Ilya Nemenman, Luís M. A. Bettencourt, Garret T. Kenyon

Red bins: Total activity histograms for all test target images. Blue bins: Total activity histograms for all test distractor images. The degree that the two distributions overlap is shown as the gray shaded area, which provides a measure of whether total luminance can be used to distinguish targets from distractors. The percentage in each shaded area shows the approximate lower bound amount of overlap of the two histograms, for comparison. Top row: Total summed activity over all retinal pixels. Little, if any bias between target and distractor images was evident in the input black and white images as there is nearly complete overlap between the distributions. Subsequent rows: Total activity histograms summed over all orientation-selective elements. Second row: Bottom-up responses prior to any lateral interactions. Third - sixth rows: Total activity histograms after - iterations of the multiplicative ODD kernel, respectively. Total summed activity became progressively more separable with additional iterations, as evinced by a decrease in the overlapping areas.


Usage metrics

PLOS Computational Biology