1989 Reports

Surface Orientation from Texture Autocorrelation

Brown, Lisa Gottesfeld

We report on a refinement of our technique for determining the orientation of a textured surface from the two-point autocorrelation function of its image. We replace our previous assumptions of isotropic texture by knowledge of the autocorrelation moment matrix of the texture when viewed head on. The orientation of a textured surface is then deduced from the effects of foreshortening on these autocorrelation moments. This technique is applied to natural images of planar textured surfaces and gives significantly improved results when applied to anisotropic textures which under the assumption of isotropy mimic the effects of projective foreshortening. The potential practicality of this method for higher level image understanding systems is discussed.