A hybrid SDF for the detailed representation of liquid-solid mixed surfaces

We propose a hybrid signed distance field (SDF) method for reconstructing the detailed surface of a model as it changes from a solid state to a liquid state. Previous particle-based fluid simulations suffer from a noisy surface problem when the particles are distributed irregularly. If a smoothing scheme is applied to reduce the problem, sharp and detailed features can be lost by over-smoothing artifacts. Our method constructs a hybrid SDF by combining level-set values from the solid and liquid parts of the object. This makes it possible to represent the detailed features and smooth surfaces of an object when both solid and liquid parts are mixed in that object. In addition, the concept of a guiding shape is proposed, which uses a coordinate-warping technique to query the level-set values quickly. The guiding shape is constructed from the object before the simulation begins and some parts of it become liquid. To track the details of the initial solid shape and preserve it, the transformation of the guiding shape is accumulated while the phase-shift is in progress. By warping the coordinates of this accumulated transformation of the guiding shape, the level-set values of the solid part can be acquired very quickly.