Fig. 1. Comparative results for isotropic and anisotropic test samples. Stochastic texture images: (a) nearly isotropic and (b) anisotropic; Gaussian ellipses: (c) nearly isotropic and (d) anisotropic; histograms of local gradient magnitudes (solid-Rayleigh model): (e) nearly isotropic and (f) anisotropic.

Fig. 2. Normal plots: (a) nearly isotropic and (b) anisotropic. Web uniformity testing results: (c) sample anisotropy ranking (our approach: solid; tensile test: dotted); (d) anisotropy measurements across the web (our approach: solid; tensile test: dotted); (e) distance profile (standard operating conditions); (f) distance profile (non-standard operating conditions).

Measured correct classification rate and computational costs: GRD, KL is our approach, i.e. bivariate Gaussian and Rayleigh models, and Kullback–Leibler distance (α = 0.15); 2G, KL is the approach of bivariate Gaussian model for the wavelet coefficients, and Kullback–Leibler distance; Gabor, Euclid. is the approach of Gabor filters, and Euclidean distance; SGLDM, Euclid. is the approach of SGLDM, and Euclidean distance; and Rayl., KL is the approach of Rayleigh model for the gradient magnitudes, and Kullback–Leibler distance