Estimation of curvature in 3D images using tensor field filtering

Abstract

This paper describes an algorithm for estimation of directionality in 2D and 3D vector fields and how that feature relates to the curvature of curves in 2D images and surfaces in 3D images. One of the main properties of the method is that no thresholding is required. It consists of two steps. First the grey level image/volume is filtered with a number of filters to obtain a tensor description of the local orientation. Secondly the tensor image/volume is filtered with a number of filters to achieve the local direction description.