3D surface texture synthesis using improved graph cuts

Abstract

This paper introduces an improved graph cut approach and combines it with commonly used surface texture representations for synthesis. Unlike graph cut which supports iterative refinement for improvement of the patch seams, the improved graph cut method re-synthesis the max error overlap region using k-nearest neighbor method. This paper follows a simple 3D surface texture synthesis framework, which comprises three steps: capture, synthesis and relighting. Firstly, it represents 3D surface textures under varied illumination directions using gradient and albedo maps. Then, it uses the surface representation to synthesis a description of a larger surface. Finally, it renders the surface representation. It proposes an improved graph cut method for the synthesis step, and the proposed method can generate perceptually smooth images of 3D surface texture from small samples. INTRDUCTIONTexture synthesis is widely recognized to be important for many applications in computer graphics, vision, and image processing. Images of real-world surface textures are affected by many properties such as object geometry and surface reflectance 18 . These surface textures are different from 2D still texture as their images can vary dramatically with illumination directions. With few exceptions, previous research into texture synthesis predominantly focuses on 2D textures 1-8 . In particular, Graph Cuts 1 is proved to be able to generate satisfying 2D textures with little seaming effects. However, 2D methods cannot provide the information required for rendering 3D surface textures with changed illumination and viewpoint conditions.There is a growing interest in the synthesis and relighting of 3D surface textures 9-23 . Most surface texture synthesis methods are direct extensions of pixel-based 12 ,21,24 or patch-based algorithms 14 ,22 . Some early approaches applied direct texture synthesis to the 3D surface 21,24 . These approaches used the Markov random field method of texture synthesis 5,7,24 which requires a spatial neighborhood function. Zalesny and Van Gool, in 16,17 present a multi-view texture model which can synthesize new viewpoints. Shum and his colleagues 20 used the CUReT database 18 to develop a method for the generation of bi-directional texture functions (BTFs). In 19 Leung and Malik proposed the use of 3D textons to synthesize new images under arbitrary viewpoints and illuminations.However, most publications on 3D surface texture are based on "image quilting" 4 or so called "smart copying" algorithms, which can produce unnatural seaming effects for surface representations.In this paper, we introduce an improved graph cut approach and combine it with commonly used surface texture representations for synthesis. The improved graph cut method re-synthesize the max error overlap region using k-nearest neighbor method. We follow a 3D surface texture synthesis framework 9 , which comprises three steps: capture, synthesis and relighting. Firstly, 3D surface textures under varied illumination directions are represented by using ...