Feature Extraction Algorithm for Banknote Textures based on Incomplete Shift Invariant Wavelet Packet Transform

Segmentation and feature extraction algorithms based on Wavelet Transform or Wavelet Packet Transform are established in pattern recognition. Especially in the field of texture analysis they are known to be practical. One difficulty of texture analysis was in the past the characterization of different printing processes. In this paper we present a new algorithmic concept to feature extraction of textures, printed by different printing techniques, without the necessity of a previous teaching phase. The typical characters of distinct printed textures are extracted by first order statistical moments of wavelet coefficients. The algorithm uses the 2D incomplete shift invariant Wavelet Packet Transform, resulting in a fast execution time of O(<i>N</i>log₂(<i>N</i>)). Since the incomplete shift invariant Wavelet Packet Transform was exclusively defined for 1D-signals, it has been modified in this research. The application describes the detection of different printed security textures.