Extraction of the basic feature points of handwriting data by auto translation error map

Abstract

Analysis of online-handwritten time series, acquired by pen tablet system, is valuable in various fields such as handwriting recognition, person verification or skill analysis. Generally, online handwriting data is a multidimensional time series comprises of time series of pen-tips position (x, y), pressure, altitude, azimuth etc.. In case of person verification application, use of such multivariate data improves verification accuracy. However increase of data volume increases computational cost for analysis. In this study, in order to reduce data volume , we propose a new method for extraction of the basic feature points of multidimensional handwriting time series from the view point of testing determinism in the underlying dynamics behind handwriting. Proposed method is based on two-dimensional recurrence map of translation error. Basic feature points denote the principal points in the trajectory of handwriting dynamics to preserve the rough form of the individual's handwriting speciality. The simulation experiment has been done with SVC 2004 online handwriting signature data. The result shows that the basic feature point series is quite sufficient for analyzing the data for identity detection while the raw handwriting time series includes redundancy.