Algorithm for Searching for Optimal Values of Parameters of Singular Point Detectors Based on Local Image Characteristics

Abstract

A method is proposed that allows one to determine the optimal values of the parameters of singular point detectors \(\mathbf{p}\) from the local characteristics of the image \({\mu}\). The method is based on searching for the extremum of the analytical dependence \(\delta=P(\mathbf{p},{\mu})\), where \(\delta\) is the detector efficiency criterion defined as the weighted sum of the type I error and type II error. The dependence \(P(\mathbf{p},{\mu})\) interpolates the set \(\delta(\mathbf{p}_{i},{\mu}_{j})\) obtained from a sample from a set of images of the same type. An experiment was performed with 40 radiographs of the lower part of the human spine, which were processed by the Harris–Stephens detector and SUSAN detector. It is found that when choosing the adjustable parameters of the detector using the proposed method, the \(\delta\) criterion turned out to be significantly lower than for the values of these parameters at the boundaries of the recommended intervals.