Support Vector Machines for Dynamic Biometric Handwriting Classification

Biometric user authentication is a recent topic in the area of computer security. This paper presents a machine learning approach to single modality user authentication. Here support vector machines (SVM) are employed to classify dynamic handwriting samples. The general goal of SVMs is to carry out binary classifications and/or to handle multiple class problems using a combination of different SVMs. Here a multi-class SVM is proposed to execute verification as well as identification of persons based on their handwriting using a given PIN and a freely chosen PIN. In the best case (trade-off for all rates) for verification using the free PIN a false acceptance rate (FAR) of 0.0083 and an attacker acceptance rate (AAR) of 0.0241 are determined while the false rejection rate (FRR) yields zero. In identification mode using the free PIN, we observe a FRR of 0.0083 and an attacker identification rate (AIR) of 0.2195 at a false identification rate (FIR) level of zero in our experiments.