Published Paper Details:

ABSTRACT: Deep learning has advanced to the point that it has surpassed old handmade methodologies and even humans for a variety of jobs in recent years. However, the amount of publicly available data for some tasks, such as the verification of handwritten signatures, is limited, making it difficult to verify the true limits of deep learning. Aside from the absence of publicly available data, evaluating the improvements of novel proposed methodologies is difficult due to the use of various databases and experimental protocols. The study's main contributions are I an in-depth analysis of state-of-the-art deep learning approaches for online signature verification; ii) the presentation and description of the new DeepSignDB online handwritten signature biometric public database; iii) the proposal of a standard experimental protocol and benchmark to be used by the research community to perform a fair comparison of novel approaches with the state-of-the-art and iv) we adopt and analyze Time-Aligned Recurrent Neural Networks (TA-RNNs), a recent deep learning approach for on-line handwritten signature verification. To train more resilient systems against forgeries, this approach combines the promise of Dynamic Time Warping with Recurrent Neural Networks. When considering experienced forging impostors and only one training signature per user, our suggested TA-RNN system beats the current state of the art, attaining outcomes even below 2.0 percent EER.