9224963 Cooper This is the first year of a three-year continuing award. Accurate, low-computational-cost recognition of 3D objects having highly- variable curved surfaces, or of the membership class of an object, or of a deformable object, have been among the research topics most resistant to satisfactory solution, and yet are necessary if reasonably general practical machine vision is to become a reality. This research explores solutions to these problems based on the unified use of patches of 3D surfaces and space curves modeled by high-degree implicit polynomials, algebraic invariants, probability distributions on the coefficient vectors for the polynomials, and asymptotic minimum probability of error (i.e., Bayesian) recognizers. The implicit polynomial curves and surfaces are generalizations of quadrics, and have much greater representation power than do quadrics or superquadrics. The algebraic invariants are functions of the polynomial coefficients having values that do not change when the data is translated, rotated or scaled differently in two different directions (i.e., an affine transformation). Among the intended applications are service robots that must recognize and handle materials and parts, service robots or machine vision systems that must interact with people, and parts inspection for manufacturing automation. The techniques to be used have received very little attention in the research literature, but now are the subject of growing interest.