Subspace method with multi scale wavelet for recognition of printer property

This paper proposes a new method of recognizing printers such as manufactures, models, and printed dates. Forensic document examiners are often asked to investigate models or manufactures of laser printers from documents. To address the difficult request, we apply subspace methods to the problem. In our previous work, we had proposed a method which extracted contours of characters as signals to conduct wavelet decomposition [1]. However, this method roughly classified models of laser printers because features used were only zero-crossing points [2]. In addition, those features were compared by only mean square error. In this paper, we use subspace method in a recognition phase and improve a feature extraction process of wavelet decomposition. The contribution of our method is demonstrated through the experiments and considerations.