A neural implementation of the JADE algorithm, called Neural-JADE, is presented and applied to natural and urban image ensembles to learn appropriate filter structures. The latter are shown to be represented quantitatively by Gabor wavelets in case of natural image stimuli and by Haar wavelets in case of urban image stimuli. Quantitative comparison concerning various filter characteristics is made with results obtained by various ICA algorithms thereby demonstrating the influence of various score functions upon the resulting filter structures. Quantitative comparison will be made also with neurophysiological characteristics of these structures.
展开▼
