Application of the Discrete Wavelet Transform to Identification of Questioned Documents

Identification of questioned documents (QDs) is the process of associating documents produced by an unknown writer with labeled archival samples. Here we use the discrete wavelet transform (DWT) to extract characteristics of letters and words of archival database samples, and questioned documents that need to be identified. Using standard clustering techniques, database document characteristics are categorized into distinct classes. These classes are labeled by the names of the known database writers. Identification of QDs is then made on the basis of the distance between QD characteristics and the database classes. We provide preliminary examples on a database of documents that was generated by nine writers. Seven different DWTs (and a voting scheme) were employed to make decisions about the QD's associations. The high success rate in identification of these samples indicates the potential that DWT-based method have in QD identification.