What question did this study set out to answer?

The study aims to develop an effective method for offline signature verification using advanced CNN techniques optimized by a grey wolf algorithm.

February 25, 2026Open Access

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

Key Points

The study aims to develop an effective method for offline signature verification using advanced CNN techniques optimized by a grey wolf algorithm.
Developed a hybrid convolutional neural network integrating residual, dense, and attention mechanisms.
Optimized network architecture using the grey wolf optimization algorithm.
Trained and tested the model on public datasets across multiple languages.
Evaluated performance on a GPU-equipped machine.
Achieved accuracy rates of 100%, 89%, and 89% for CEDAR, BHSig260-Hindi, and BHSig260-Bengali datasets.
Demonstrated quick response times of 0.0066, 0.0046, and 0.0066 seconds for respective datasets.
GWO significantly reduced the effort required to find optimal learning parameters.

Abstract

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.

Bookmark

View Full Paper