An inverse rendering approach for heterogeneous translucent materials

Since heterogeneous translucent materials, such as natural jades and marble, are complex hybrids of different materials, it is difficult to set precise optical parameters for subsurface scattering model as the material really has. In this paper, an inverse rendering approach is presented for heterogeneous translucent materials from a single input photograph. Given one photograph with an object of a certain heterogeneous translucent material, our approach can generate material distribution and estimate heterogeneous optical parameters to render images that look similar to the input photograph. We initialize material distribution using 3D Simplex Noise combined with Fractal Brownian Motion, and set color pattern of the noise using histogram matching method. The volume data with heterogeneous optical parameters is initialized based on the value of color pattern matched noise, and it is rendered in a certain lighting condition using Monte Carlo ray marching method. An iteration process is designed to approximate optical parameters to minimize the difference between rendering result and input photograph. Then the volume data with optimal heterogeneous optical parameters is obtained, which can be used for rendering any geometry model in different lighting conditions. Experimental results show that heterogeneous translucent objects can be rendered precisely similar to the material in the photograph with our approach.