A robust discrete curvature estimator was recently proposed by Kerautret et al. [1]. In this paper, we exploit the precision and sta- bility of this estimator in order to define a contour extraction method for analysing geometric features. We propose to use a reference curva- ture function for extracting the frontier of a shape in a gray level image. The frontier extraction is done by using both geometric information rep- resented by the reference curvature and gradient information contained in the source image. The application of this work is done in a medical application.