NONLINEAR MULTISCALE REPRESENTATIONS FOR IMAGE SEGMENTATION

In order to segment an image the use of information at multiple scales is invaluable. The hyperstack, a linking-model-based segmentation tec hnique, uses intensity to link points in adjacent levels of a scale sp ace stack. This approach has been successfully applied to linear multi scale representations. Multiscale representions which satisfy two scal e space properties, viz. a causality criterion and a semigroup propert y in differential form, are valid inputs as well. In this paper we con sider linear scale space, gradient-dependent diffusion, and the Euclid ean shortening flow. Since no global scale parameter is available in t he latter two approaches we compare scale levels based on evolution ti me, information theoretic measures, and by counting the number of obje cts. The multiscale representations are compared with respect to their performance in image segmentation tasks on test and MR images. The hy perstack proves to be rather insensitive to the underlying multiscale representation although the nonlinear representations reduced the numb er of post processing steps. (C) 1997 Academic Press.