Stroke Point Warping-Based Reference Selection and Verification of Online Signature

Abstract

Signatures alignment and reference selection are an important task for signature verification. Due to inherent variability of the acquired signature, a novel technique called stroke point warping (SPW) is proposed. After incorporating the SPW technique, the normalized correlation coefficient of two signatures from the same signer is improved. A novel reference selection strategy is also proposed by combining the SPW method and the normalization of signature support. Two functional features based on shape signature and instantaneous phase are introduced. The corresponding global features, i.e., entropy of the shape signature and the first two statistical moments of instantaneous phase are also derived. Further, a novel fusion technique is proposed in this paper, where the global features (derived from dynamic signature profile) are combined with functional reference signature-based scores. In the signature verification framework, the number of sample signatures available for a person is small compared with the available features. The minimum redundancy maximum relevance method is applied to rank the features. After feature selection, the support vector machine-based verification is evaluated. The proposed algorithm, evaluated on the task 2 SVC2004 database, achieves the equal error rate of 1%, which is better than the state-of-the-art algorithms.