Flexible Painting-Based Volume Classification Using Ellipsoid Gaussian Transfer Function

Abstract

The paper focuses on appearance and spatial property-based volume classification with a painting interface. Two key problems exist, i.e., effectively propagating user-given painting to the entire volume and intuitively controlling the roles of appearance and spatial properties, which are not well solved in existing painting-based transfer function specification methods. The present paper formulates painting propagation as a function interpolation problem in a high-dimensional affine space and solves it effectively using Gaussian radial basis functions. For the second problem, the present work presents a two-step approach, which first propagates the user-given painting to the entire volume using appearance- and spatial-property-dominated feature vectors, and then combines the painting propagation results using an ellipsoid Gaussian transfer function (ETF) for volume classification. The user can intuitively manipulate ETF using system-provided widgets. The effectiveness of the proposed method has been verified on several datasets.