Synthesis-by-analysis of complex textures

A technique for unsupervised texture synthesis by analysis is presented. It is based on a stochastic approximation of a textured field obtained by nonlinearly transforming a complex white Gaussian random field. The nonlinear transformation is constituted by two linear filters connected by a complex hard-limiter. The identification of the texture model is performed by means of a Bussgang blind deconvolution algorithm exploiting a generalization to the complex case of the Van Vleck rule. After a theoretical discussion of the method typical examples are provided.