SELECTION OF CRITERIA FROM STANDARD DESCRIPTIONS OF NORMAL 2-DIMENSIONAL SETS

We scatter of criteria (signals) within a given class is represented i n the form of an ellipse of equal probability density functions. A met hod is described for determining the coefficients of the equation of t he ellipse from the results of experimental studies of the criteria in certain specimens of one class with variable uncontrolled parameters. We probabilistic measure of the informativeness of a pair of criteria is the condition of external tangency of the ellipses of two adjacent classes with a homothetic increase in the dimensions of the ellipses relative to their centers. We informativeness of the criteria is exami ned relative to the dimensions and positions of the ellipses, and desc riptions are given of algorithms for classification and measurement ba sed on a pair of criteria. We method used to choose the most informati ve criterion pairs is substantiated to enhance the reliability of imag e identification, classification, or measurement.