<p>Signatures have for quite some time been viewed as the most acknowledged and pragmatic methods for client confirmation, notwithstanding being powerless against talented falsifiers. As the signature is the essential instrument for confirmation and approval in legitimate exchanges, it becomes necessary to perform signature verification. This paper proposes a pre-trained EfficientNetB7 model-based Siamese neural network (eSNN) for writer-independent offline signature verification (WIOfSV) system. The EfficientNetB7 model is used for feature extraction in the twin network of the Siamese neural network. Unlike the previous approaches, this approach does not rely on hand-crafted feature engineering; instead, it learns its features from data in a writer-independent scenario. The eSNN method is robust, enhances feature extraction and reduces the training complexity of the Siamese neural network by using the pre-trained weights on ImageNet. The performance of well-known ConvNets employed as feature extractors in the sub-network of the Siamese neural network is compared. The performance of the eSNN method is evaluated using different evaluation parameters for comparison on six other public datasets. The experimental analysis shows that the eSNN achieves state-of-the-art or competitive performance on five out of six datasets.</p>