Procedure for generating template masks for detecting variable signals

摘要

This paper is concerned with the use of template masks for locating objects and their features in digital images. In an ideal case suitable masks would take the form of spatial matched filters for the chosen object or feature profile. However, in practical situations, not only will noise be present but also the signals themselves may vary. This paper analyses the situation and formulates a procedure for generating suitable masks in such cases. The overall filter can be construed as a special type of spatial matched filter with integral noise-whitening. The method is applied to edge, corner and hole detection, and tradeoff formulae are derived connecting accuracy, sensitivity and computational load, with number of masks as the single free parameter.