A dynamic vision classification system using Fourier descriptions

The design and implementation of a vision classifier system which can display, label, and identify different known objects in applicable images is presented. Every object can be described during the training phase or classified during the recognition phase according to the normalized Fourier descriptors obtained from sampling its boundary. These descriptors are invariant to the rotation, size, and orientation of the corresponding object. The system allows the user to dynamically train and store new patterns in the appropriate knowledge base of patterns. It is implemented on an AI VAXstation. The descriptors provide an increasingly accurate characterization of shape as more coefficients are included. Each descriptor is a measure of the lobedness of the subject. The descriptors provide accurate classifications for all objects tested under moderate noise and small distortion