The main purpose of the digitalSnow project is to provide efficient computational tools to study the snow metamorphism from 3D images of real snow microstructures acquired using X tomography techniques. We design 3D image-based numerical models than can simulate the shape evolution of the snow microstructure during its metamorphism. The feature extraction from a form has been widely studied in the literature : many algorithms are dealing with meshes, point clouds, or discrete data. We propose a new method with theoretical guarantees to detect features on digital data by analyzing, in scale-space, recent digital curvature estimators with good mathematical properties. In our previous work, we proposed curvature estimators (mean, principal, Gaussian) dealing with 2d and 3d digitized shapes where we have demonstrated multigrid convergence (under certain conditions on the form), depending on a parameter : the radius of integration. We propose to study these curvature estimators with as parameter the scale-space of the integration radius, for a given digital form. By analyzing the behavior of these estimators, we can extract the features (singularities, edges, linear parts, constant curvature, etc.) of the digitized form. We use this classification for digital snow microstructure analysis (e.g. to detect to crystallographic orientations or bonds between snow grains ).