Human Signatures are still used in banks, organizations, and in many other security issues. Currently, the need not to touch any physical components, minerals, or devices has become a very important necessity, especially in the spread of viruses that are transmitted and largely preserved in minerals. This paper presents a new algorithm to identify and verify humans based on the enrolled signatures. Some features may influence the shape, rotation, and structure of the digital signature. All these features should be taken into consideration as it may be varied randomly every time each person enrolled the signature to the system. In this paper, we took three important Region of Interest (RoI) named as Multi-Region of Interest (MRoI) by which most common features of the entered signatures are taken into consideration. The MRoI are equal splitted region that are convoluted to produce one template applied to support vector machine (SVM) classifier. Every RoI of the signature are then applied to local binary pattern (LBP) feature extractor, then it convoluted to one template pattern to be classified using SVM. Furthermore, Deep Convolutional Neural Networks (DCNN) is presented for both feature extraction and classification stages to boost the results obtained for MRoI using SVM. We present fully connected layer of DCNN for 128 person, Further, we implement the proposed architecture using dropout softmax based on SVM. The proposed system is designed to handle both Arabic and English handwritten signatures collected from 128 individuals and the accuracy achieved is 95%.
展开▼
