High discriminant features for writer-independent online signature verification

The application of online signature is promising. However, its huge intra-individual variability and its extremely low inter-class distance brought by forged signatures make it difficult for even state-of-the-art online signature algorithms to be applied in practical scenarios on a large scale. This paper proposes a semantic-driven extraction method of high discriminative features for writer-independent online signature verification, addressing the problem of representation learning with high discriminative features. The semantic-driven model aims at learning the high-level semantic representation of the writer's inherent signature habits, and it has combined the advantages of LSTM and CNN. Furthermore, several global feature descriptors are designed to extract writer habitual features such as speed, and writing pressure at keystroke positions. The most difficult, writer-independent, 1v1 experiments on the three benchmark data sets of MCYT-100, SUSIG, and MOBISIG were performed, and the results show that the performance of the proposed method is better than that of the state-of-the-art methods, and its performance on the MCYT-100 dataset is 16% higher than the second-best method.