Online Signature Verification Based on Recursive Subset Training
In this paper, a novel approach has been proposed for online signature verification based on recursive subset training. Our approach is based on estimating the Equal Error Rate (EER) of the entire system and then splitting the entire data set into two subsets based on the EER of the system. The two subsets includes writers whose individual EER is more than the EER of the system and writers whose EER is less than the EER of the system. This procedure is recursively repeated until writer level parameters are decided. Unlike other verification models where same features are used for all writers, our approach is based on identifying writer dependent features and also writer dependent thresholds. Initially, writer dependent features are selected using a suitable feature selection method. Signatures are clustered using Fuzzy C means and represented in the form of interval valued symbolic feature vector. Signature verification is done based on the selected representation and the EER of system is calculated. Once the EER of the system is estimated, our method is based on estimating the EER of individual writers and splitting the dataset into subsets and estimating the EER of each of the subset separately. This process of splitting the dataset into subset and treating each of the subsystem separately is repeated until the individual writer thresholds and features are identified. We conducted experiments on MCYT-DB1 to show the effectiveness of our novel approach.
KeywordsSubset recursive training writer dependent parameters Feature selection Fuzzy C means Symbolic feature vector
Unable to display preview. Download preview PDF.
- 5.Fiérrez-Aguilar, J., Krawczyk, S., Ortega-Garcia, J., Jain, A.K.: Fusion of Local and Regional Approaches for On-Line Signature Verification. In: Li, S.Z., Sun, Z., Tan, T., Pankanti, S., Chollet, G., Zhang, D. (eds.) IWBRS 2005. LNCS, vol. 3781, pp. 188–196. Springer, Heidelberg (2005)CrossRefGoogle Scholar
- 6.Plamondon, R., Lorette, G.: Automatic signature verification and writer identification: the state of the art. Pattern Recognition 2(2), 63–94 (1989)Google Scholar
- 12.Parodi, M., Gomez’, J.C., Liwicki, M.: Online Signature Verification Based on Legendre Series Representation. Robustness Assessment of Different Feature Combinations. In: International Conference on Frontiers in Handwriting Recognition (ICFHR 2012), pp. 715–723 (2012)Google Scholar
- 15.Chen, Y., Ding, X.: Online signature verification using direction sequence string matching. In: Proceedings of SPIE, vol. 4875, pp. 744–749 (2002)Google Scholar
- 16.Wang, K., Wang, Y., Zhang, Z.: On-line Signature Verification Using Segment-to-segment Graph Matching. In: International Conference on Document Analysis and Recognition (ICDAR 2011), pp. 805–808 (2011)Google Scholar
- 17.Guru, D.S., Manjunatha, K.S., Manjunath, S.: User dependent features in online signature verification. In: Swamy, P.P., Guru, D.S. (eds.) ICMCCA 2012. LNEE, vol. 213, pp. 229–240. Springer, Heidelberg (2013)Google Scholar
- 20.Kira, K., Rendell, L.: A practical approach to feature selection. In: Proceedings of 9th International Workshop on Machine Learning, pp. 249–256 (1992)Google Scholar
- 22.Cai, D., Zhang, C., He, X.: Unsupervised Feature Selection for Multi-cluster Data. In: 16th ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD 2010), pp. 333–342 (2010)Google Scholar
- 23.Guru, D.S., Prakash, H.N., Manjunath, S.: Online Signature Verification: An approach based on Cluster Representation of Global Features. In: Seventh International Conference on Advances in Pattern Recognition, pp. 209–212 (2009)Google Scholar