Button to scroll to the top of the page.

Events

Qualifier: Anqi Zhang
Thursday, November 05, 2020, 12:00pm

Anqi Zhang, UT-Austin

"The Human Visual System Whitens in Space But Not in Spatial Frequency"

Abstract: Detecting object targets in natural backgrounds is a significant natural task for human and primate visual systems. Natural backgrounds (and medical image backgrounds) typically modulate in contrast over space, and have amplitude spectra that fall inversely with spatial frequency. The effect of these two properties of natural backgrounds can be studied by measuring and modeling thresholds for identifying targets in contrast-modulated 1/f noise. In this task, the ideal observer computes the response of a template (“receptive field” in the shape of the target) that is weighted (whitened) in the Fourier domain by the inverse of the power spectrum, and is weighted (whitened) in the space domain by the inverse of the local contrast power. Whitening in spatial frequency capitalizes on the variations in the background’s amplitude spectrum across spatial frequency; whitening in space capitalizes on the variations in the background’s contrast across space. Suboptimal models include applying (a) a template that is whitened neither in spatial frequency nor space (the ideal observer for detection in white noise), (b) a template that is only whitened in spatial frequency (the ideal observer for detection in 1/f noise), and (c) a template that is only whitened in space (the ideal observer for detection in modulated white noise). In simulations, we determined the identification thresholds of the optimal and suboptimal models for spatially-windowed periodic targets that varied in spatial frequency content (sine wave, triangle wave, square wave, and rectangle wave). Whitening in spatial frequency is least beneficial for the sine wave target, and most beneficial for the rectangle wave, and this benefit holds for uniform and contrast modulated 1/f noise. Whitening in space is equally beneficial for all targets, but only in modulated noise. Measurements on human subjects are most consistent with model (c), where the template is only whitened in space.

Location: Zoom