TEXTURE SYNTHESIS USING BI-DIRECTIONAL TEXTURE FUNCTIONS FOR REAL-WORLD SURFACES

摘要

This paper presents an effective approach for generating textures from real-world surfaces using an image-based technique built on the Bidirectional Texture Functions (BTFs). A BTF characterizes the appearance of a surface at fine scale in terms of the viewing and lighting directions in addition to the two dimensional texture. Our approach generates textures using a sparse set of images captured from different viewing and lighting directions to achieve photorealistic appearance at mesostructural level. Taking a sparse set of reference BTF images, the first step normalizes the multiple texture databases to allow uniform mapping procedure. The second step performs distance metric searching for proper images in the sparse set, having the closest viewing and lighting setting as the three dimensional geometric scene. Finally, a block-matching algorithm is taken to preserve the details of appearance and to recover the color of the texture. The whole process uses image-based techniques. It directly synthesizes a texture on the surface of the 3D model to be rendered. The experimental results demonstrate the effectiveness of our approach.