Tradeoffs between speed and accuracy in two-stage template matching

Abstract This paper examines the tradeoff between speed and accuracy in two-stage template matching. A criterion function based approach shows that on some occasions random sampling of the input image data leads to an unexpected improvement over systematic skimming. Formulae are obtained which provide practical tests for optimisation. They should be useful in a wide variety of applications and are tested here in the context of edge detection in grey-scale images.