Applications of time-series analysis: A case study on the development of a predictive model for handwritten signature

Abstract

Handwritten signature represents a person’s identity. Although overall patterns among the signatures of same person remain same, there can appear natural variations because two or more signatures of same person written within a moment and keeping a sufficient time gap, cannot be exactly same. These natural variations result in intrapersonal variations. In the present study, signature samples were collected from each participant under different situations of body position, paper texture, paper position etc. to successfully capture the intrapersonal variations. Two features, namely area and height-width ratio (HWR) were extracted for each signature using appropriate image processing techniques. These features were then modelled to the Single Exponential Smoothing Time Series Technique as well as our developed methodology to predict the variations. Using this technique the Positive Predictive Values (PPV) and False Rejection Rate (FRR) for both these features were found to be 88% , 12% and 95.78%, 4.22% respectively.

Keywords

• offline signature analysis
• image processing
• time series
• single exponential smoothing technique
• predictive modeling