Online signature verification system with anti-forgery provision based on segmentation and structure learning of HMM
Dapeng Zhang, Shinkichi Inagaki, Naoki Kanada, Tatsuya Suzuki · 2010
Inspired by forensic experts working on authentication of oriental characters like Chinese and Japanese who usually rely on distinguishing detailed features of individual strokes such as dots and straight lines, our new HMM model consisted of many sub-models each represents an individual stroke of a signature. Furthermore, 3 models were compared in 2 steps using a hierarchical manner. First, original user was distinguished from data corpus consisted of random forgeries; Secondly, original user was distinguished from skilled forgeries.