Efficient Set-D4: A Deep Learning Approach to Identify Human Handwritten Signatures
Shailesh Sahu, Sudhakar D. Khamankar, Amit Kumar Tripathi · 2024
Identification of handwritten signatures has experienced extensive usage in the world of information handling. Likewise, there are numerous distinctions in people’s styles of writing; it can be difficult to correctly recognise those individuals from photographs. This process is also complicated by the existence of numerous visual artifacts, such as vibration, deformation and intensity variations. This paper proposed a deep learning (DL)-based approach to categorising integers termed EfficientDet-D4 using the recommended method in an attempt to overcome those restrictions. To clearly show the region of fascination, the input signature images are first accurately annotated. These photos are used to train the Efficient Set-D4 algorithms to detect and categorise their unique signatures. Utilising the MNIST dataset evaluate the detection accuracy of the suggested model and succeeded in obtaining a 99.83% total accuracy rating. Furthermore conducted the cross-dataset analysis on the USPS datasets and received a dependability score of 99.10%. The graphical outputs and the real-world results show that system can accurately identify handwritten signatures compared to pictures regardless of how their writing style differs, as well as when there are many testing artifacts there, including motion, deception, chrominance, status fluctuations and dimension shifts of numerals.