Self-Optimizing, Self-Learning System in Pictorial Pattern Recognition

Publisher Website
Google Scholar
Abstract
A system of computer programs discriminates between pictorial patterns by determining a substantial number of numerically encoded pattern properties. Supervised learning is used to find both an optimum decision sequence and the thresholds for decision rules. These are applied to patterns from an object set to test the consistency of the classification procedure. Nonsupervised learning is used in the pattern detection section of the program. The system has been extensively tested in the discrimination of biomedical patterns from their digitized microscopic images, specifically in the machine recognition of tumor cells and of cells involved in immune response. Objective cytodiagnostic decision making was shown to be more consistent, and, in certain instances, capable of finer discrimination than assessment by qualified human observers.

This publication has 3 references indexed in Scilit: