A multistage algorithm for fast classification of patterns

Abstract A multistage statistical pattern classification algorithm is proposed. The algorithm consists of three consecutive stages: (1) parallelpiped classification, (2) a new method for ellipsoidal separation, (3) Mahalanobis minimum distance classification. The multistage classifier is designed such that points not classified by a given stage are considered by the next one. The performance of the classifier is tested using a synthetic image. It has been found that this approach reduces computer classification time at a reasonable expense of classification accuracy. The algorithm performs well for the classification of remote sensing images and is implemented on a microcomputer.