Dynamic Biometric Recognition of Handwritten Digits Using Symbolic Aggregate Approximation

Abstract

Symbolic aggregate approximation (SAX) is an ideal technique for dynamic biometric recognition of handwritten digits. The manipulation of time series in SAX readily lends itself to analysis of the spatial coordinate data acquired from a digit written on the touchscreen of a smartphone or tablet. SAX generates a sequence of alphabetic characters derived from a time series as a result of this analysis. Alphabetic sequences may be compared using the SAX minimum distance function. We propose a new algorithm for author authentication based on this process and the simple use of mean and standard deviation. We analyze the accuracy of our solution using JMotif, a Java time series data mining toolkit based on SAX, and a handwritten digit database of 1400 samples from 14 authors. Our experimental validation proves that our algorithm will authenticate the author of any handwritten digit almost 100% of the time. We conclude that our work has important implications in the design of handwritten Personal Identification Number systems.