A new class of classification and thresholding algorithms for image processing based on entropy-related measures

A parametric mutual certainty measure is introduced, which is closely related to entropy. By relating mutual certainty to the structure of decision trees as well as to the Bayesian probability of error, a hierachical classification algorithm is developed, which can be used for the classification of digital imagery on the basis of their feature domain. By means of the parameter values the characteristics of the decision tree can be influenced. Furthermore, in the case of some stopping criterion the partitioning until then guarantees a minimization of the probability of error on the average. The algorithm can be used in itself or in combination with other algorithms, e.g. local refined boundaries in the case of application of `box'-classifiers. Furthermore, by taking the inverse of the decision criterion, an algorithm is obtained which is useful for gray level thresholding of digital imagery. Attention is paid to the observed duality of classification in the feature space and gray level thresholding of histograms