Gradient threshold selection using the facet model

摘要

Automatic gradient threshold selection for edge detection is a non-trivial task due to the presence of image noise. This problem is posed within a statistical framework based on a cubic facet model for the image data and a Gaussian model for the noise. Under these assumptions, two statistics which are functions of the gradient strength and facet residual error are derived. Experiments show that thresholds on these statistics produce results which are superior to those obtained by the best subjective threshold on the gradient image. A Bayes decision procedure is developed which makes threshold selection automatic.