Premium
Rough sets in hybrid methods for pattern recognition
Author(s) -
Cyran Krzysztof A.,
Mrózek Adam
Publication year - 2001
Publication title -
international journal of intelligent systems
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.291
H-Index - 87
eISSN - 1098-111X
pISSN - 0884-8173
DOI - 10.1002/1098-111x(200102)16:2<149::aid-int10>3.0.co;2-s
Subject(s) - pattern recognition (psychology) , computer science , artificial intelligence , feature vector , rough set , feature extraction , artificial neural network , invariant (physics) , measure (data warehouse) , data mining , mathematics , mathematical physics
The article shows how rough sets can be applied to improve the classification ability of a hybrid pattern recognition system. The system presented here consists of a feature extractor based on a computer‐generated hologram (CGH) playing the role of a ring‐wedge detector. Features extracted by it are shift, rotation, and scale invariant. Although they can be optimized, no method has been proposed in the literature. This article presents an original method of optimizing the feature extraction abilities of a CGH. The method uses rough set theory (RST) to measure the amount of essential information contained in the feature vector. This measure is used to define an objective function in the optimization process. Since RST‐based factors are not differentiable, we use a nongradient approach for a search in the space of possible solutions. Finally, RST is used to determine decision rules for the classification of feature vectors. The alternative method of classification based on neural networks is also discussed. The whole method is illustrated by a system recognizing the class of speckle pattern images indicating the class of distortion of optical fibers. © 2001 John Wiley & Sons, Inc.