Image Analysis Using Mathematical Morphology
IEEE Transactions on Pattern Analysis and Machine Intelligence
Automatic generation of morphological set recognition algorithms
Automatic generation of morphological set recognition algorithms
Performance-driven autonomous design of pattern-recognition systems
Applied Artificial Intelligence - Special issue: design for high autonomy
System Identification through Simulated Evolution: A Machine Learning Approach to Modeling
System Identification through Simulated Evolution: A Machine Learning Approach to Modeling
Genetic Algorithms in Search, Optimization and Machine Learning
Genetic Algorithms in Search, Optimization and Machine Learning
Machine learning procedures for generating image domain feature detectors
Machine learning procedures for generating image domain feature detectors
Image Analysis and Mathematical Morphology
Image Analysis and Mathematical Morphology
| Hi-index | 0.01 |
The E-Morph learning algorithm combines a number of learning algorithms-genetic, evolutionary programming, clustering-into a hybrid learning system for solving multiclass pattern-recognition problems. Our work also shows that a randomly generated pool of primitive detectors, rather than manually coded features, can be enhanced and assembled into effective solution sets.The design of automatic pattern recognition programs by humans is a difficult problem requiring deduction, intuition, and experience. This is particularly true for the development of recognition systems based on mathematical morphology. (See the accompanying box on the next page for an explanation of morphological representation.) This article discusses the techniques we have investigated for using evolutionary learning to generate morphological pattern recognition systems.Automating this learning process requires certain intellectual information to initialize it. This information falls naturally into three basic elements: a suitable problem representation a search strategy, and performance measures A representation includes the basic functions, operators, and data structures that effectively characterize the solution space or possible range of attainable solutions. Search strategies define the rules or techniques for manipulating the representation and navigating the search space.The search strategy discussed here is based on evolutionary learning. The primary performance measure is generally defined in terms of the final solution-recognition accuracy. As the complexity of problems increases, additional performance measures may be needed to guide the learning through the intermediate phases of system design and to confront credit assignment problems.