Differential Filtering and Detexturing

Abstract

Extracting valuable information from 2D or 3D visual data plays an important role in image and geometry processing. Surfaces obtained through a scanning process or other reconstruction algorithms are inevitably noisy due to error in the scanning process and resampling of the data at various processing steps. These surfaces need to be denoised both for aesthetic reasons and for further geometry processing. Similarly, extracting or removing texture patterns from 2D or 3D data is challenging due to the complication of its statistical features. In this dissertation, I describe how to remove surface noise and image texture patterns. In particular, I focus on denoising triangulated models based on L0 minimization, in which a very important discrete differential operator for arbitrary triangle meshes has been developed. Compared to other anisotropic denoising algorithms, our method is more robust than other anisotropic denoising algorithms, and produces good results even in the presence of high noise. I also introduce how to use bilateral filter appropriately on image texture removal by modifying its range image. While current existing methods either fail to remove the textures completely or over blur main structures, our method delivers best-in-class image detexturing performance.