Extrapolation of bidirectional texture functions using texture synthesis guided by photometric normals

Abstract

Numerous applications in computer graphics and beyond benefit from accurate models for the visual appearance of real-world materials. Data-driven models like photographically acquired bidirectional texture functions (BTFs) suffer from limited sample sizes enforced by the common assumption of far-field illumination. Several materials like leather, structured wallpapers or wood contain structural elements on scales not captured by typical BTF measurements. We propose a method extending recent research by Steinhausen et al. to extrapolate BTFs for large-scale material samples from a measured and compressed BTF for a small fraction of the material sample, guided by a set of constraints. We propose combining color constraints with surface descriptors similar to normal maps as part of the constraints guiding the extrapolation process. This helps narrowing down the search space for suitable ABRDFs per texel to a large extent. To acquire surface descriptors for nearly flat materials, we build upon the idea of photometrically estimating normals. Inspired by recent work by Pan and Skala, we obtain images of the sample in four different rotations with an off-the-shelf flatbed scanner and derive surface curvature information from these. Furthermore, we simplify the extrapolation process by using a pixel-based texture synthesis scheme, reaching computational efficiency similar to texture optimization.

Neural Photometry-guided Visual Attribute Transfer

“…Inspired by such methods, we leverage several images of the material under different illumination conditions, however, we require less data than typical BTFs capture setups [35], [36]. Similarly, the problem of extrapolating BTFs captures to larger material samples was addressed by Steinhausen et al [37], [38], who propagate measured BTFs using texture synthesis. Our method is not meant to propagate full BTFs measurements but could potentially be applied to such datasets, as we illustrate on the supplementary material.…”