Customized data filtering for mobile signature verification

Abstract

Many studies for signature verification focus on the improvement of verification accuracy, the most significant issue. They makes subject models with sufficient sample signatures since the verification accuracy is generally proportional to the number of sample data. However, in the mobile environment, excessive requests of signatures could exhaust users to abandon the corresponding services. This paper therefore proposes customized signature filtering scheme that reduces the number of signature requests with preserving the verification accuracy. On the basis of the observation that large variations between sample signatures disturb fast and precise learning of a classifier, the proposed scheme filters out unnecessary samples. In other words, it compares sample signatures, eliminates out-of-range signatures, and uses only others as the sample data. The experiment results show that the proposed scheme reduces 51% signature requests without decrease of the verification accuracy. © Research India Publications.