Abstract:
Identification of individuals from handwritten documents using automated recognition systems has gained significant research interest due to the wide variety of applications it offers for forensic analysis, signature verification, classification of historical writings and other document analysis tasks. In this paper, we present a framework that combines different feature space representations of handwriting for an effective characterization of writers. Multiple distance functions are applied to each feature space which are then combined to enhance the overall recognition performance. The proposed identification framework evaluated on a standard database realizes significant performance improvements in terms of identification rate.
