Pattern Classification

Classification • Supervised – parallelpiped – minimum distance – maximum likelihood (Bayes Rule) > non-parametric > parametric – support vector machines – neural networks – context classification • Unsupervised (clustering) – K-Means – ISODATA • Pattern recognition in remote sensing has been based on the intuitive notion that pixels belonging to the same class should have similar gray values in a given band. – Given two spectral bands, pixels from the same class plotted in a two-dimensional histogram should appear as a localized cluster. – If n images, each in a different spectral band, are available, pixels from the same class should form a localized cluster in n-space.