Writer Identification through Information Retrieval: The Allograph Weight Vector

We show a number of promising results in writer identification, by recasting the traditional information retrieval (IR) problem of finding documents based on the frequency of occurrence of their terms. In IR, the tf-idf is a well-known statistical measure that weighs the importance of certain terms occurring in a database of documents. Here, writers are searched on the basis of the frequency of occurrence of particular character shapes: the allographs. The results show a high retrieval score. Moreover, by using the af-iwf (allograph frequency inverse writer frequency) measure, qualitative and quantitative analyses can be made that elaborate on the particular allograph shapes that lead to a successful writer identification. In this paper, we sketch the application of these techniques in forensic science.