Volume Denoising for Visualizing Refraction

Rendering refraction in volume visualization requires smoothly distributed normals to synthesize good quality visual representations. Such refractive visualizations are more susceptible to noise in the data than visualizations that do not involve refraction. In this paper, we present the first study to investigate convolution and diffusion methods with respect to this problem, highlighting the need for improving the smoothness of dataset normals while minimizing unnecessary geometrical distortion. We outline the mathematical specifications of four volume denoising filters, namely (a) Gaussian convolution, (b) inhomogeneous linear diffusion, (c) anisotropic nonlinear diffusion and (d) regularized anisotropic nonlinear diffusion. We apply these filters to volumetric scenes involving refractive objects, and introduce two metrics for measuring distortion and smoothing effects. Our results indicate that it is possible to make significant improvements in image quality in refractive visualization without introducing excessive distortion.