Reduced complexity modeling and reproduction of colored textures

An unsupervised color texture synthesis-by-analysis method is described. Th e texture is reproduced to appear perceptually similar to a given prototype by copying its statistical properties up to the second order. The synthesi zed texture is obtained at the output of a Single-Input Three-Output nonlin ear system driven by a realization da white Gaussian random field. Signific ant complexity reduction is gained by exploiting the rank deficiency of the Cross Power Spectral Density Matrix of the color texture samples.