Pattern Classification: A Unified View of Statistical and Neural Approaches

Based on many years of practical experience predominately in the area of character recognition and document analysis, the author details a number of competing approaches to building up essential estimating functions--statistical modeling, least mean squares techniques and radial basis functions. Traditional statistics-based pattern classification techniques as well as connectionist and neural methods are coherently treated and shown to be inextricably interfused. Uses an extremely simplified two-dimensional example task throughout the text to illustrate diverse approaches in a unified manner.