Writer Identification Using an HMM-Based Handwriting Recognition System: To Normalize the Input or Not?

A system for writer identification based on handwritten text lines is described in this paper. The system uses Hidden Markov Model based recognizers which are designed for text line recognition. Features are extracted from a text line and used to train the recognizers. Prior to feature extraction, normalization operations are applied to a text line. On the one hand, there exists a strong correlation between the text recognition and the writer identification rate, and applying normalization operations increases the text recognition rate. On the other hand, normalization also removes writer-specific information from a handwritten text line. Hence there is a trade-off between optimizing the text recognition performance of our system and keeping writer specific features. In this paper, we study the effect of normalization operations, such as slant correction, width normalization, and vertical scaling, on the identification rate of our system.