Department of Pattern Recognition

(eng): We inspect the ability to reproduce spectral (color) composition in random field-based texture models, test when it can neglect spectral correlation, and simplify these random models without visibly depreciating their visual quality. These probabilistic models present essential two or three-dimensional factors for modeling seven-dimensional Bidirectional Texture Function (BTF) - the most advanced representation currently used in real-world material visual properties modeling. They can seamlessly approximate original measured massive data and extend them to an arbitrary size or simulate unmeasured ones. Using extensive test data sets and a small set of setup control parameters, these models reach a vast compression ratio while maintaining the visual quality of measurements, and thus, they are the only viable alternative for BTF practical usage.