Part of Advances in Neural Information Processing Systems 9 (NIPS 1996)
Stephen Grossberg, James Williamson
A self-organizing architecture is developed for image region classi(cid:173) fication. The system consists of a preprocessor that utilizes multi(cid:173) scale filtering, competition, cooperation, and diffusion to compute a vector of image boundary and surface properties, notably texture and brightness properties. This vector inputs to a system that incrementally learns noisy multidimensional mappings and their probabilities. The architecture is applied to difficult real-world image classification problems, including classification of synthet(cid:173) ic aperture radar and natural texture images, and outperforms a recent state-of-the-art system at classifying natural textures.