Evolution of Morphological Recognition Systems

E-MORPH is a multi-phase evolutionary learning system that evolves cooperative sets of feature detectors and combines their responses using a nearest neighbor classifier to form a complete pattern recognition system. Detectors are represented as sequences of morphological operators that perform a complex series of image-to-image transformations resulting in the localization of information. E-MORPH has been applied to several image analysis tasks including optical character recognition (OCR) and automated target recognition (ATR). This paper describes the application of E-MORPH to a problem in the area of medical image analysis. Results from the analysis of a set of x-rays of human spinal columns are presented. The goal of this study is to evolve a morphological program that locates individual vertebra in an x-ray of a patient's spine and ultimately measures the level of deterioration of the spinal column.