Offline Signature Forgery Detection Using Image Processing

Fingerprints, irises, faces, voices, and handwritten signatures are five prevalent biometric recognition in many practical fields such as financial payment, attendance,… Signatures are used above all others as a primary form of authentication for a range of transactions. Detecting forged signatures is a crucial task that presents several challenges. The obstacles in handwritten signature verification include large variations within the genuine signatures of the same person, small differences between genuine and skillfully forged signatures. Manual hand-written signature verification's efficiency tends to be inconsistent based on expertise of the experts. As a result, manual signature review can lead to an uncomfortable number of false rejections and forgery acceptances. In this paper, a solution based on Convolutional Neural Network using pre-trained VGG16 model is proposed for offline handwritten signature automatical verification system. The proposed model obtains high accuracy multi-class classification with a few training signature samples. Images are pre-processed using a series of image processing techniques to separate the signature pixels from the background and noise pixels. To avoid overfitting, the dataset is divided into 3 groups for training and testing. During training process, the model is trained with 2 methods: without fine-tuning and fine-tuning the pre-trained layers. The proposed models, which are trained with 2 types methods, are tested with the same dataset and the accuracies are compared for finding which training type gives the best result. It is very encourage that the model work effectively when using fine-tuning the pre-trained layers with accuracy attains the figure 97.71%. It shows that the method is efficient and can be applied to large datasets of signatures.