Skeleton Extraction via Anisotropic Heat Flow

Skeleton is a thin version of the shape, which is a useful feature for shape description in image processing and computer vision. In skeleton extraction, it is an important task to compute a smooth and high quality medial function that can provide skeleton features easily and without noise. One of the main features for skeleton point detection is curvature maxima along the level-sets of the medial function. We present the proposed anisotropic heat diffusion as a novel medial function generation technique, which can remove noise in the image and can preserve the prominent curvatures of the object shape along the level-sets. The proposed anisotropic heat diffusion equation is obtained with the following considerations. Let η denote the direction normal to the feature boundary (the gradient direction) through a given point in an image I, and let τ denote the tangent direction. These directions can be written in terms of the first derivatives of the image, Ix and Iy, as