Verification of Offline Signature Based on a Hybrid Dense-Attention-Residual-CNN Structure Optimized by Grey Wolf Algorithm

Signature verification is a necessary vision task with prevalent use in securing realworld applications. Offline signature is a popular type that needs to efficacious approach to be checked due to the depending on the paper time and the written tool. The progress in intelligent algorithms participates in supporting simple way authentication-based signatures. In this research work, an offline signature verification (OSV) based on an optimized and costumed convolutional neural network(CNN) infrastructure is suggested. The CNN is costumed by integrating residual, dense, and attention mechanisms, then optimized by applying a grey wolf optimization(GWO) algorithm. The system is trained and tested with public datasets with different languages, then conducted with a machine equipped with a suitable graphic processing unit(GPU). The assessment results show a remarkable accuracy of 100%, 89%, and 89% over a testing time of 0.0066, 0.0046, and 0.0066 sec for CEDAR, BHSig260- Hindi, and BHSig260- Bengali, samples. In addition, the utilization of the GWO algorithm reduced the designer's effort to get the best learning parameters. Therefore, the suggested system can be considered a balanced automated design in classifying the signature images as genuine or forged. In addition, this work can support making a smart and authenticated tool.