Local-feature-based similarity measure for stochastic resonance in visual perception of spatially structured images

Abstract

For images, stochastic resonance or useful-noise effects have previously been assessed with low-level pixel-based information measures. Such measures are not sensitive to coherent spatial structures usually existing in images. As a result, we show that such measures are not sufficient to properly account for stochastic resonance occurring in visual perception. We introduce higher-level similarity measures, inspired from visual perception, and based on local feature descriptors of scale invariant feature transform (SIFT) type. We demonstrate that such SIFT-based measures allow for an assessment of stochastic resonance that matches the visual perception of images with spatial structures. Constructive action of noise is registered in this way with both additive noise and multiplicative speckle noise. Speckle noise, with its grainy appearance, is particularly prone to introducing spurious spatial structures in images, and the stochastic resonance visually perceived and quantitatively assessed with SIFT-based measures is specially examined in this context.

Tuning the Noise in Magnetic Resonance Imaging to Maximize Nonlinear Information Transmission

“…This choice affects second-order statistics of the noise, and therefore has no influence on the measures of similarity which are based on the first-order statistics of the noise. However, the study of the influence of the sampling strategy in the k-space would be an interesting perspective especially in the direction of a very recent work [40] which demonstrates the influence of the second-order statistics in the visual perception of stochastic resonance.…”

Computer vision tools to optimize reconstruction parameters in x-ray in-line phase tomography

“…For each keypoint, a description vector is computed through measures in the spatial neighborhood of the gradient. The number of pairs of keypoints with similar description vectors, called SIFT matches, are here, and as in Delahaies et al (2012), are used as a measure of local similarity between the atlas image and the sample image. The matching procedure includes two filtering steps.…”