The influence of the γ-parameter on feature detection with automatic scale selection

Abstract

A method to automatically select locally appropriate scales for feature detection, proposed by Lindeberg [8], [9], involves choosing a so-called γ-parameter. The implications of the choice of γ-parameter are studied and it is demonstrated that different values of γ can lead to qualitatively different features being detected. As an example the range of γ-values is determined such that a second derivative of Gaussian filter kernel detects ridges but not edges. Some results of this relatively simple ridge detector are shown for two-dimensional images.