Spatial Pyramid Matching-based Multi-script Off-line Signature Identification

Among all of the biometric authentication systems, handwritten signatures are considered as the most legally and socially accepted attributes for personal identification. The objective of this investigation is to present an empirical contribution towards the understanding of a signature identification technique involving multi-script off-line signatures. In our experiment, SIFT (Scale-Invariant Feature Transform) descriptors with Spatial Pyramid Matching (SPM)-based approaches have been used for feature extraction of signatures written in multiple scripts. Support Vector Machines (SVMs) are employed as the classifier in this experiment. 300 classes from the publicly available GPDS[16] dataset consisting of 7200 (300 × 24; 24 signature samples in each class) genuine signatures, 300 classes from a Devnagari signature dataset consisting of 7200 (300 × 24) genuine signatures and 200 classes from Bangla signature dataset consisting of 4800 (200 × 24) genuine signatures have been considered for this experiment. The signature identification experiment is conducted on these three datasets separately as well as 800 classes from a combined dataset of English, Devnagari, and Bangla signatures. The identification accuracy on the datasets is encouraging and 99.32% accuracy was obtained on the combined dataset of signatures, while 99.95%, 99.25% and 99.57% accuracy were achieved on experiments conducted separately on English, Devnagari, and Bangla signature datasets.