Shape from local and global analysis of texture

Abstract

We demonstrate methods for estimating shape from texture, for textures which are neither isotropic nor homogeneous. Whereas earlier work assumed either isotropy or homogeneity, we make a more general assumption: the local tangent distribution of the surface texture is invariant with respect to position. We then present a taxonomy of texture types, and a set of methods that can be used to recover surface orientation for each texture type. We define two classes of methods for estimating shape from texture: invariance-seeking methods, and value-seeking methods. These definitions are used to account for the performance of new methods on different classes of texture. The approach outlined in this paper provides a general strategy for generating methods for obtaining shape from texture for known classes of texture.