Welcome to the Bahria University DSpace digital repository. DSpace is a digital service that collects, preserves, and distributes digital material. Repositories are important tools for preserving an organization's legacy; they facilitate digital preservation and scholarly communication.
dc.contributor.author | 03-243201-012, USMAN AMJAD | |
dc.date.accessioned | 2023-02-07T08:37:25Z | |
dc.date.available | 2023-02-07T08:37:25Z | |
dc.date.issued | 2022-10-10 | |
dc.identifier.other | BULC999 | |
dc.identifier.uri | http://hdl.handle.net/123456789/14857 | |
dc.description | Principal Supervisor: Dr Iram Noreen | en_US |
dc.description.abstract | Handwritten signature verification has got community attention among numerous biometric systems during this decade. It is widely used as identification and verification of a person, transaction or document in organizations, banks, law courts, business processes. Offline signatures are mere images of signatures and are mostly managed by Computer Vision techniques such as template matching, and statistical methods. Recently, Hidden Markov models and Neural Networks based approaches are used in the problem domain. However, despite wide-ranging work by research community signature verification still remains open to the research due to the diverse challenges such as intra class variability among signature of same individual. Further, extraction of discriminative visual features is another challenge while using machine learning approaches. Deep learning approaches are not highly explored in this domain due to non-availability of large datasets due to privacy restrictions. This study aims to propose a latent deep learning based approach for automatic signature verification to be used on mobiles and other less resourceful devices In our approach we have explored an adhoc MobileNet-V2 to learn weights using offline triplet loss. SVM, Random Forest, MLP and Adaboost classifiers are used to generate result, the best being Random Forest on Bengali handwritten signature dataset. I am able to achieve 86% accuracy score with skilled forgeries | en_US |
dc.language.iso | en | en_US |
dc.relation.ispartofseries | ;BULC999 | |
dc.title | OFFLINE MODE AUTOMATED SIGNATURE VERIFICATION | en_US |
dc.type | Project Reports | en_US |