Automated Signature Authentication Using Deep Convolutional Neural Network

Verifying the authenticity of handwritten signatures remains a vital challenge in biometric security systems, especially in financial, legal, and administrative domains. This work presents a deep learning–based framework for automated offline signature authentication using transfer learning with MobileNetV2. The system distinguishes real from forged signatures through convolutional feature extraction and classification. Leveraging Mobile NetV2s lightweight architecture and ImageNet pretraining, the model is fine-tuned on a labelled signature dataset using data augmentation and adaptive training. The pipeline consists of four core modules: Data Preprocessing, CNN Feature Extraction, Transfer Learning Fine-Tuning, and Real-Time Signature Prediction. Evaluation on a real-world dataset achieves 96.46 parcent accuracy, as shown in the confusion matrix. This framework provides a reliable, scalable solution for signature verification in practical security applications, especially where computational efficiency and high precision are essential.