DeepSign: Deep On-Line Signature Verification

Deep learning has become a breathtaking technology in the last years, overcoming traditional handcrafted approaches and even humans for many different tasks. However, some tasks, such as verification of handwritten signatures, amount publicly available data is scarce, what makes difficult to test real limits deep learning. In addition lack public data, it not easy evaluate improvements novel proposed databases experimental protocols are usually considered. The main contributions this study are: i) we provide an in-depth analysis state-of-the-art on-line signature verification, ii) present describe new DeepSignDB biometric database, iii) propose standard protocol benchmark be used research community order perform fair comparison with state art, iv) adapt our recent approach named Time-Aligned Recurrent Neural Networks (TA-RNNs) task verification. This combines potential Dynamic Time Warping train more robust systems against forgeries. Our TA-RNN system outperforms achieving results below 2.0% EER when considering skilled forgery impostors just one training per user.