Handwritten signature verification using shallow convolutional neural network

Handwritten signatures are an undeniable and unique way to prove the identity of persons. Owing to the simplicity and uniqueness, it finds an essential place in the area of behavioral biometric. Signatures are the most widely accepted biometric trait by law enforcement agencies/personnel for verification purposes, especially in financial institutions, legal transactions, etc. and hence secured authentication becomes imperative. In the era of a digital age, numerous transactions are taking place, where handwritten signature verification is required by the agencies, e.g., banks, etc. In such scenarios, the process of signature verification, besides being accurate and secure, should be very fast, i.e., the real-time verification can be done. In this paper, we have proposed a convolutional neural network-based language-independent shallow architecture (sCNN(Shallow Convolutional Neural Network)) for signature verification. The proposed architecture is very simple but extremely efficient in terms of accuracy. A custom shallow convolution neural network is used to automatically learn the features of signature from the provided training data. Another contribution of the research work, which is the handwritten signature data collection for 137 subjects and 467 subjects, which are named as CVBLSig-V1 and CVBLSig-V2 respectively, has been reported in this paper. The performance of the proposed architecture has been evaluated on publicly available datasets, i.e., MCYT-75, MCYT-100, and GPDS, as well as CVBLSig-V1 and CVBLSig-V2. The performance was also compared with state of the art reported methods and shown improved, while considering the accuracy and equal error rate (EER) as performance metrics.