Biometric Signature Recognition Using DTW and Fractional Distances

Distance metrics have long been fundamental tools in machine learning, crucial for assessing similarity between data points within an n-dimensional space. These metrics play a significant role in pattern recognition applications, including online signature biometric recognition. Traditionally, standard distance measures like the Euclidean distance have been employed for this purpose. However, recent advancements highlight the need for more sophisticated distance metrics that can better capture the intricacies of datasets and improve learning outcomes. In this paper, we delve deeper into the impact of different distance metrics-Absolute, Euclidean, Quadratic, and p-norm-on the accuracy of biometric online signature recognition. Specifically, we employ Dynamic Time Warping to classify signatures using a cost matrix derived from these metrics at each sampling point, represented by an eight-dimensional vector encompassing spatial coordinates and pressure from a Wacom digitizing tablet, alongside their first and second derivatives. Our analysis sheds light on the significance of the p-value in influencing recognition accuracies, offering valuable insights for the advancement of biometric signature recognition systems.