Automatic Analysis of Handwriting for Demographic Classification of Writers (P-0128) (MFN 5497)

Abstract

Handwriting and analysis of handwriting has been an active and interesting area of research for many decades. Despite the growth of digital media, handwriting retains its importance as being one of the very fine motor skills that contributes to the intellectual development of individuals as well. Computerized analysis of handwriting finds applications in areas like handwriting recognition, signature verification, classification of ancient manuscripts, forensic applications and authentication of writer of a document. Anohter interesting aspect of handwriting is the existence of correlation between writing and the different demographic attributes of the writer. Although such studies have been carried out in psychological sciences for many decades now, automation of this analysis through computer programs is a relatively recent development. The system developed in our study predicts the demographic attributes of an individual through offline images of handwriting. The focus of our study lies on two such attributes, gender and handedness. The methodology is based on extracting a set of textural features from writing samples of different writers and training a support vector machine classifier to learn the different demographic classes. During evaluation phase, the features of the query writing sample are fed to the trained classifier which outputs the class labels. The system trained and evaluated on a benchmark database of handwritten samples (QUWI) realized promising results.