Tests of Gaussianity and linearity for random fields using estimated higher order spectra

摘要

We present frequency domain tests of Gaussianity and linearity for random fields and study their power both analytically and by simulation. Our tests conform to the likelihood ratio principle, and the test statistics (chi-squared in both cases) are simple and intuitively plausible. Using these tests, we found certain texture images to be both non-Gaussian and nonlinear.