Estimating Scattering Parameters for Surfaces with Different Curvature using Directional Dipole Model
ZENG XINGJI (1551131)
Acquisition of parameters for the Bidirectional Scattering Surface Reflectance Distribution Function (BSSRDF) has significant meanings in realistic image synthesis. In this thesis, we aim to estimate scattering parameters from non-planar surface and present an inverse rendering approach combining a optimization algorithm with a newly developed BSSRDF, directional dipole model, for parameter estimation.Being aware that the surface curvature has a noticeable affection on the visual effect of translucency, we turn to this often ignored factor to improve the accuracy of our approach. We assume that surface curvature also affects the estimation results and design experiments to verify our assumption.As a beginning, we make a explicit evaluation for the optimization algorithm and the directional dipole model. For evaluation of the optimization algorithm, parameters are estimated from observed data generated by directional dipole model and the algorithm finds the ground truth exactly. As for valuation of directional dipole model, parameters are estimated from simulated sphere, real planar. A comparison with dipole model is also done. Experimental results show directional dipole model is suitable for estimation in spherical surface.
To figure out how surface curvature affects the estimation results, parameters are estimated from both simulated and real spheres of different radius. According to the estimation results, higher error tends to appear in extremely small(high curvature) and extremely large(low curvature) spheres.Therefore, we can draw a conclusion that it is necessary to consider surface curvature to improve the accuracy of inverse rendering.Note that we limit our target to sphere for simplicity in this thesis, though, theoretically our approach can also be applied to other non-planar surface.