Handwritten Signature Verification Model Using Transfer Deep Learning Technique

The legitimacy and authentication of handwritten signature verification in industries such as banking, security and educational institutions cannot be over-emphasized. Handwritten signatures are physical representations of a person’s signature that can be used to sign documents. This research work designed a model for verifying handwritten signatures using transfer deep learning which involved the use of a Visual Geometry Group (VGG16), a 16-layer deep pre-trained algorithm and a custom CNN. A total of 14,500 signatures were harvested from attendance registers of staff from selected higher institutions in the western part of Nigeria. The data were pre-processed to extract distinct features of the handwritten signatures before the classification and verification processes. During classification, the technique exploited the knowledge gained from reference signatures using distinct threshold values to match the new signatures for proper verification. The results from the experiment demonstrated a good upshot with performance accuracy of 99%, precision of 99%, recall of 100%, and f-score of 100%. Other evaluation metrics used stand at good performance values respectively. This implied that the technique can accurately classify and authenticate the reference features. The model can therefore be used for efficient verification of handwritten signatures in our institutions.