A Hierarchical Approach for the Offline Handwritten Signature Recognition

The domain of offline handwritten signature recognition deals with establishing the owner of a signature in areas where a person's identity is required. As the amount of handwritten signatures is constantly increasing, it becomes harder to distinguish among the signature instances, that is why methods should be found in order to maintain a good separation between the signatures. Our work is focused on identifying the techniques that could be employed when dealing with large volumes of data. In order to achieve this goal, we propose a hierarchical partitioning of data by utilizing two dataset reduction techniques (feature selection and clustering) and by finding the classifier that is appropriate for each signature model. By applying our proposed approach on a real dataset, we report the best results for a dataset of 14 instances/class divided into 8 clusters and a recognition accuracy of 92.11%.