Direct curvature scale space in corner detection

Abstract

Curvature Scale Space (CSS) representation of planar curves is considered to be a modern tool in image processing and shape analysis. Direct Curvature Scale Space (DCSS) is defined as CSS that results from convolving the curvature of a curve with a Gaussian kernel directly. Recently a theory of DCSS in corner detection has been established. In the present paper the DCSS theory is considered to transform the DCSS image of a given curve into a tree organization, and then corners on the curve are detected and located in a multiscale sense. Experiments are conducted to show that the DCSS corner detector can work equally well as the CSS corner detector does on curves with multiple-size features, however, at much less computational cost. © Springer-Verlag Berlin Heidelberg 2006.