A set of novel features for writer identification

Abstract

A system for writer identification is described in this paper. It first segments a given page of handwritten text into individual lines and then extracts a set of features from each line. These features are subsequently used in a k-nearest-neighbor classifier that compares the feature vector extracted from a given input text to a number of prototype vectors coming from writers with known identity. The proposed method has been tested on a database holding pages of handwritten text produced by 50 writers. On this database a recognition rate of about 90% has been achieved using a single line of handwritten text as input. The recognition rate is increased to almost 100% if a whole page of text is provided to the system. © Springer-Verlag 2003.