The accurate validation of signed documents is important to ensuring personal privacy and digital safety and security. Offline handwritten signature recognition is now widely used in sectors like banking, healthcare, and legal proceedings. However, there remains a security risk in that a handwritten signature might easily be forged by a malicious third party. There is an urgent need in many sectors to improve the current recognition techniques so that they can identify faked autographs.
Research from China published in the International Journal of Biometrics has introduced a new approach to offline handwritten signature recognition based on Generative Adversarial Networks (GANs). This approach allows variable features such as a measure of the pen pressure and tilt angle to be used for signature recognition. Xiaoguang Jiang of the Department of Culture and Arts at Yongcheng Vocational College explains that by integrating GANs to enhance accuracy it is possible to generate realistic virtual signatures that alongside training classifiers from authentic written signatures and so improve the accuracy of the classifier.
Jianh has used a Deep Convolutional GAN (DCGAN) model and demonstrated 95 percent accuracy in tests, which is much greater than the accuracy possible with earlier models. Accurate signature recognition is critical for identity verification processes in finance, law, healthcare, and other areas. The same techniques might also be applied more esoterically to signature verification in the art world, for instance, or the digitization of historical documents to ensure authenticity and provenance.
Jiang, X. (2024) ‘Offline handwritten signature recognition based on generative adversarial networks’, Int. J. Biometrics, Vol. 16, Nos. 3/4, pp.236–255.