Handwritten Signature Verification Using Convolution Neural Network (CNN)

Abstract

Despite recent widespread research in the field, handwritten signature verification is still an unresolved research problem. A person’s signature is an important biometric trait of a human that can be used to verify a person’s identification. There are two primary biometric identification methods: (i) A method of identification based on vision and (ii) An identification method without the use of vision. Examples of vision-based identification include face reading, fingerprint identification, and retina scanning. The other examples for non-vision-based identification include speech recognition and signature verification. In financial, commercial, and legal activities, signatures are crucial. Two methods are widely studied and investigated for signature verification: the online method (dynamic method) and the offline method (Static approach). Offline systems are more practical and user-friendly than online systems, but because they lack dynamic information, offline verification is regarded to be more difficult. Systems for verifying signatures are designed to determine if a particular signature is authentic (made by the claimed individual) or a forgery (produced by an impostor). The data collection, feature extraction, feature selection, and classification model make up the bulk of the suggested model. A convolutional neural network is used to extract features, and machine learning algorithms are used to verify handwritten signatures. To train CNN models for feature extraction and data augmentation, raw images of signatures are employed. VGG16, Inception-v3, Res-Net50, and Xception CNN architectures are employed. The recovered attributes are classified as authentic or false using Euclidean distance, cosine similarity, and supervised learning techniques such as Logistic Regression, Random Forest, SVM, and its variants. Data from ICDAR 2011, including pairwise-organized Signature Datasets, was used for testing. The database comprises the signatures of 69 different people.