On-line handwritten signature verification using hidden Markov model features

HMM-based on-line signature verification: Feature extraction and signature modeling

“…, N. (Rabiner, 1989). Some previous works using HMMs for signature verification include (Yang et al, 1995;Kashi et al, 1997;Dolfing et al, 1998). Basically, the HMM represents a doubly stochastic process governed by an underlying Markov chain with finite number of states and a set of random functions each of which is associated with the output observation of one state (Yang et al, 1995).…”

“…In regional methods, the time functions are converted to a sequence of vectors describing regional properties. One of the most popular regional approaches is the method based on Hidden Markov Models (Yang et al, 1995;Kashi et al, 1997;Dolfing et al, 1998). In most of these cases, the HMMs modeled stroke-based sequences.…”

Fusion of Local and Regional Approaches for On-Line Signature Verification

“…Examples can be found regarding other biometrics like fingerprint, where the combined use of local (e.g., minutiae-based) and global (e.g, ridge-based) approaches has been shown to improve verification performance [16], or speaker verification, where multilevel approaches are currently the state-of-the-art [17]. Some work on multi-level signature verification is available in [5][6][7].…”

“…In regional methods, the time functions are converted to a sequence of vectors describing regional properties. One of the most popular regional approaches is the method based on Hidden Markov Models (HMM) [5,7,8,12].…”

“…Different approaches have been considered in the literature in order to extract relevant information from on-line signature data [3]; they can broadly be divided into: i) feature-based approaches, in which a holistic vector representation consisting of a set of global features is derived from the signature trajectories [4][5][6][7], and ii) function-based approaches, in which time sequences describing local properties of the signature are used for recognition (e.g., position trajectory, velocity, acceleration, force, or pressure) [5][6][7][8][9][10][11][12]. Although recent work shows that feature-based approaches are competitive with respect to function-based methods in some conditions [7], the latter methods have traditionally yielded better results [3].…”