Statistical Approaches to Material Classification.
Manik Varma, Andrew Zisserman · 2002
The objective of this paper is classification of materials from a single image obtained under unknown viewpoint and illumination conditions. Texture classification under such general conditions is an extremely challenging task. Our methods are based on the statistical distribution of rotationally invariant filter responses in a low dimensional space. There are two points of novelty: first, two representations of filter outputs, textons and binned histograms, are shown to be equivalent; second, two classification methodologies, nearest neighbour matching and Bayesian classification, are compared. In essence, given the equivalence of texton and bin representations, the paper carries out an exact comparison between the texton based distribution comparison classifiers