Higher-orderstructureof naturalimages

We present a statistical model for learning efficient codes of higher-order structure in natural images. The model, a non-linear generalization of in- dependent component analysis, replaces the standard assumption of inde- pendence for the joint distribution of coefficients with a distribution that is adapted to the variance structure of the coefficients of an efficient im- age basis. This offers anovel description of higher order image structure and provides a way to learn coarse-coded, sparse-distributed represen- tations of abstract image properties such as object location, scale, and texture.