Unsupervised Selective Anisotropic Feature Diffusion for Cortical Surface Registration
In this paper we use the theory of geometric curve and surface evolution to diffuse an object preserving their important features. We do that using a curvature flow method with a novel influence function. The proposed influence function takes profit from the characteristics of the level set representation to perform a directional diffusion improving other influence functions results that evolve in such a way that though they diffuse noise, important features are also lost after some iterations. We show that if the geometry of the object is important for our purposes our technique outperforms the classical ones.