Super Resolution of 3D Surface Texture

Abstract

Real-world surface textures commonly consist of a spatial variation of reflectance properties combined with rough surface geometry, called 3D surface texture. Variation of illumination can produce dramatic changes their appearance. While previous work on super resolution pays great attention to 2D images, this paper proposes an inexpensive method for super resolution on 3D surface textures. First, four images under different illumination angles are captured for each sample surface texture, from which the surface gradient maps and the albedo map are extracted by Photometric Stereo. Then the surface gradient and albedo maps and three selected original images, which represent 3D surface texture, are processed by a modified 2D Example-based super resolution method to generate result representation sets with higher resolution. Finally, relight surface representations to generate high resolution texture images under any different illumination conditions. Experimental results show that this method provides more fine details compared to interpolation techniques.