Salience le orientation-filter response measured as suspicious coincidence in natural images
Conference Paper
Overview
Identity
Additional Document Info
View All
Overview
abstract
Visual cortex neurons have receptive fields resembling oriented bandpass filters, and their response distributions on natural images are non-Gaussian. Inspired by this, we previously showed that comparing the response distribution to normal distribution with the same variance gives a good thresholding criterion for detecting salient levels of edginess in images. However, (1) the results were based on comparison with human data, thus, an objective, quantitative performance measure was not taken. Furthermore, (2) why a normal distribution would serve as a good baseline was not investigated in full. In this paper, we first conduct a quantitative analysis of the normal-distribution baseline, using artificial images that closely mimic the statistics of natural images. Since in these artificial images, we can control and obtain the exact saliency information, the performance of the thresholding algorithm can be measured objectively. We then interpret the issue of the normal distribution being an effective baseline for thresholding, under the general concept of suspicious coincidence proposed by Barlow. It turns out that salience defined our way can be understood as a deviation from the unsuspicious baseline. Our results show that the response distribution on white-noise images (where there is no structure, thus zero salience and nothing suspicious) has a near-Gaussian distribution. We then show that the response threshold directly calculated from the response distribution to white-noise images closely matches that of humans, providing further support for the analysis. In sum, our results and analysis show an intimate relationship among subjective perceptual measure of salience, objective measures of salience using normal distributions as a baseline, and the theory of suspicious coincidence. Copyright 2006, American Association for Artificial Intelligence (www.aaai.org). All rights reserved.
