Feature Extraction for Quantitative Classification of Marbles

In this paper, our purpose is to find and test several features that can be used for classification of marble slabs. The database, which consists of 193 marble cubes, is classified into four groups by the experts. In this classification, it is observed that the experts pay careful attention to texture, homogeneity and the color of the cube surfaces. To represent these properties, Co-occurence matrices and sum and difference histograms are used to extract seven features. Dimension of the feature space is reduced by using Principle Component Analysis before discussing the performances of above mentioned feature extraction methods. Finally, the effect of color space on classification performance is investigated.