MULTIDIMENSIONAL COOCCURRENCE MATRICES FOR OBJECT RECOGNITION AND MATCHING

A novel method is proposed for object recognition and matching. It is based on the automatic search of features that characterize a certain object class using a training set consisting of both positive and nega tive examples. Special multidimensional co-occurrence matrices are use d for the description and representation of some basic image structure s. The features are extracted from the elements of this matrix and exp ress quantitatively the relative abundance of some elementary structur es, i.e., they are quotients of certain elements of the matrix. Only f eatures which discriminate the classes clearly are used. The method is demonstrated in numerous applications, falling under the general prob lems of texture recognition, texture defect detection, and shape recog nition. (C) 1996 Academic Press, Inc.