Multilingual-Signature Verification by Verifier Fusion Using Random Forests

In this paper, we propose an improvement to the method of combined segmentation verification for multi-script signature verification. The proposed method employs a fusion strategy for multiple signature verifiers using different modalities, i.e., offline and online signature verification. We use offline signature shape features extracted from separated three color plane (RGB) images that reflect the pen pressure and pen velocity of the signature signers. The Mahalanobis distance for each offline feature vector is calculated for signature verification. In addition, we employ another offline feature based on the grayscale histogram and similarity between histograms for offline signature verification. The online feature-based technique employs a dynamic programming matching technique for the time series data of the signatures. These matching results are fused using a verification classifier for making the final decision. Conventionally, a support vector machine (SVM) has been used for the verification classifier. We investigate the performance and feasibility of a random forest (RF) for the verification classifier instead. The results of evaluation experiments using the SigComp multi-script signature dataset show that the proposed method improves the performance and that RF outperforms SVM.