Estimation of Textured Surface Inclination by Parallel Local Spectral Analysis

When an inclined, uniformly textured surface is viewed by an observer or imaged by a camera, the systematic distortions of the perspective transformation will induce a predictable distribution of shifts in the projected spatial frequencies which compose the texture. By measuring these shifts using a set of filters having suitable spatial, frequency, and orientation resolution, the inclination angles of the original textured surface may be estimated. An algorithm is presented which uses the amplitude distributions of 2D Gabor filters to perform such a calculation. Central to the algorithm is a pair of iteratively executed routines. The fist adjusts local sets of parameters to reduce the error between predicted and measured filter amplitudes. The second propagates the local parameters to neighboring regions to consolidate the estimates of inclination. The algorithm is capable of operating in parallel on any number of regions in the image and with a diverse set of filter inputs.