Off-Line Signature Confirmation based on Cluster Representations of Geometrical and Statistical Features through Vector Distance, Neural Network and Support Vector Machine Classifiers

Abstract

We exploited the geometrical and statistical properties of signature images for offline signature verification and identification in this paper, using signature clustering and classification based on extracted features. The Offline-SVR has been tested on the 2004 Ministerio de Ciencia Tecnología e Innovación (MCTYTDB) OffLineSignSubCorpus dataset, the MCYT-330 online signature dataset, and the MCE-200 dataset, which together are referred to as the MCE-605 dataset. Using a standard data set for experiments, the results of the Vector Distance (VD), Support Vector Machine (SVM) and Neural Network (NN) methods are significantly superior to those of other signature verification and recognition methods. Moreover, the VD method performed better than The SVM and NN methods. The purpose of the study is on clustering signature images using geometric and statistical features, as well as the utilization vector distance, neural networks, and support vector machines for signature image verification and identification. It was decided to use the algorithm for developing geometric and statistical features. The signature images are classified using generated features using k-means clustering, and Offline and Online- Support Vector Regression (SVR) is accomplished using VD, SVM, and NN training and classification with a different number of signatures each time, preceded by verification using recognition statistics. Because of the minimal number of features, the designed mechanism seems to be much faster. Experimenting on a standard dataset reveals that the results obtained from clustering signatures and categorization are effective and simple in comparison to other Offline signature confirmation systems. In this research work, we address the problem of representing handwritten signatures (online/offline) suitable for effective verification and recognition. We propose effective feature extraction for verification and recognition of signatures.