Automatic Classification of Paper Using Combinational Optimization of Image Features

抄録

For security printings such as banknotes and securities, the ability to recognize fakes and forgeries is important. However, adding new features to security printings is not easy because it leads to cost increases. We therefore focused on information in the paper of such printed matter. The information of paper is changed by conditions of production. Therefore, it is conceivable that information of the paper is effective in the classification of the paper. Previous studies proposed, as a method of paper classification, a focus on features of paper images, such as periodic wire marks, and this has been shown to be effective. Hence we considered paper classification methods that reflected changes in a greater variety of features, and found that combining multiple features computable from images of the paper was effective.
This report describes a method of automatic classification of paper that used a Support Vector Machine which selects, with a Genetic Algorithm, multiple image features computable from captured images of papers. The image features used for the classification are features calculated from the pixel values, features obtainable from the binary image, features obtainable from the segmented image, and features computable from the Gray Level Co-occurrence Matrix. Experiments have been carried out to determine the ability of the proposed method. Five kinds of paper provided by different manufacturers, and hand-made paper made under different conditions, were examined as specimens in these experiments to automatically configure of a paper classifier.
The results showed that, using the proposed method, it was possible to automatically configure a precision paper classifier capable of recognizing specific papers with different characteristics.