Abstract
This paper presents a novel method called PNLICA for image extraction from nonlinear mixtures of mutually independent images. Post nonlinear mixtures (PNL) is used for modelling the mixing process. A modified multilayer perceptron (MLP) is combined with a higher order statistics linear independent component analysis (ICA) model to sequentially extract the hidden images one-by-one from the PNL mixtures. A kurtosis-based unsupervised learning algorithm is used to adapt the model. Through computer simulation, it is observed that the proposed model is capable of effectively separating the source images from only the knowledge of nonlinear mixture of sources.