Diffusion Tensor Weighted Harmonic Fields for Feature Classification

Abstract

This paper presents an efficient method for feature definition and classification on shapes. We tackle this challenge by exploring the weighted harmonic field (WHF), which is also the stable state of a heat diffusion regulated by an anisotropic diffusion tensor. The technical merit of our method is highlighted by the elegant integration of locallydefined diffusion tensor and globally-defined harmonic field in an anisotropic manner. At the computational front, the partial differential equation of heat diffusion becomes a linear system with Dirichlet boundary condition at heat sources (also called seeds). We develop an algorithm for automatic seed selection, enhanced by a fast update procedure in a high dimensional space. Various experiments are conducted to demonstrate the ease of manipulation and high performance of our method.