Pattern Recognition Methods for Texture Analysis Case Study: Steel Surface Classification

The problem of studying pattern recognition techniques for analyzing textured surfaces is considered in this thesis and the results are applied to the classification of steel surfaces according to their surface properties. Various texture analysis techniques are studied and features are extracted from steel surfaces. Two new texture analysis methods are introduced and tested. To simplify and enhance the classification operation, only representative features extracted from the steel surfaces are selected by feature selection algorithms. Two new feature selection algorithms are introduced and are tested. Relative performances of feature selection algorithms are also tested on the features obtained. For this reason a performance measure for feature selection algorithms is introduced. Selected features by various feature selection algorithms are fed into classifiers to discriminate between different classes. To test the performances of different classification algorithms on the selected features, various classification algorithms are used. The majority voting technique is also tested for combining the results of various classifiers.