Object Recognition from a Perimeter High Curvature Model using Rank Conditioned Morphological Operators

Abstract

To reduce the processing overhead, recognition algorithms require some simpler description of the picture under test, and the objects it may contain. The paper describes a solution to the recognition problem using rank conditioned morphological operators. Recognition of a given model pose from incomplete data is made possible by using a “best fit” approximation based on the rank measure extracted for that pose. An objective measure of the degree of confidence in the recognition result is generated by assessing the degree of occupancy of the model set by the data under test. A structural feature based approach with multiple pose library models is used to identify objects belonging to the (necessarily) limited world of the system. The extracted data consists of a set of descriptors calculated from the positions of the points of high curvature about the perimeter of the object.