Super Resolution of 3D Surface Texture Based on Eigen Images

Abstract

Real-world surface texture generally comprises rough geometry and various reflectance properties, which can dramatically change its appearance under different illumination directions, namely 3D surface texture. Thus, traditional methods for super resolution of 2D texture images cannot be directly used to generate 3D surface texture with higher resolution. This paper presents an efficient method for super resolutionon 3D surface texture. It focuses on synthesizing real-world surface textures with higher resolution and under varied illumination directions. To this extent, it first employs an Eigen-based method to represent 3D surface texture images under multiple illumination directions. Then it modifies the examplebased super resolution method proposed by Freemanet al. to generate new 3D surface texture. Experimental results based on a set of real-world surface textures are presented and show the effectiveness of the new algorithm.