Writer-independent online handwritten signature verification using stable segments via an autoencoder-based Siamese network
Fangjun Luan, Cheng Zhu Lu, Shuai Yuan · 2025
To enhance the accuracy of online handwritten signature verification, a writer-independent online handwritten signature verification system is proposed using stable segments via an autoencoder based Siamese network. During the online signature verification process, we first extract different stroke segment sets from each user signature as candidates for the stable segment set, which includes two steps: segment alignment via a window cumulative difference matrix and segment set extraction using a multi-distance dynamic time warping similarity calculation algorithm. An autoencoder and Siamese network are then combined to encode and classify the different segment sets for signature verification. The method was validated on the SVC-2004-task2 dataset and the MCYT-100 dataset. Experimental results represented that when selecting the top 65% of segment similarities after segment matching as stable segments, the optimal performance of signature verification is achieved. This method achieved authentication rates of 90.63% and 94.40%, and equal error rates of 9.75% and 6.20%, respectively. These experimental results indicate that the proposed method has certain advantages in the effectiveness of signature verification.