Statistical significance of features in digital images

摘要

This paper develops a methodology for finding which features in a noisy image are strong enough to be distinguished from background noise. It is based on scale-space, i.e. a family of smooths of the image. Pixel locations having statistically significant gradient and/or curvature are highlighted by colored symbols. The gradient version is enhanced by displaying regions of significance with streamlines. The usefulness of the new methodology is illustrated by the analysis of simulated and real images.