January 10, 2016 -
Biometrics Research Group, Inc. defines signature recognition as a behavioural biometric that identifies an individual on the basis of their handwritten text.
Writing is human physical expression but concurrently an acquired skill. Signature recognition requires an individual to supply a sample of text which serves as a base of measurement of their writing. The purpose of the signature recognition process is to identify the writer of a given sample, while the purpose of a signature verification process is to confirm or reject the sample. Writing sample can be examined by way of two separate techniques.
The first technique is static. It requires the individual to supply their signature on paper, where it will be digitized through an optical scanner or camera. The data, in turn, run through a software algorithm that recognizes the text by way of analyzing its shape. The this technique is referred to as an “off-line” mode of recognition.
Off-line handwriting recognition is an important form of biometric identification because signatures are a socially accepted identification method which are commonly used for bank, credit card and various business transactions. Off-line signature processing are typically used in office automation systems that validate cheques, credit cards, contracts and historical documents.
Static, off-line handwriting recognition is performed after a text sample has been completed and digitally captured. The optically captured image data is then converted into a bit pattern. Off-line signature processing has total nearly 40 features including analysis of center of gravity, edges, and curves for authentication. Off-line signature recognition can thus be a challenging task due to normal variability in signatures and the fact that dynamic information regarding the pen path is not available. Moreover, sample data is normally limited to only a small number of signatures per individual. Shape matching is normally treated by determining and matching key points so as to avoid the problems associated with the detection and parameterization of curves.
The first technique for signature recognition is dynamic. Dynamic signature recognition is a biometric modality that uses, for recognition purposes, the anatomic and behavioral characteristics that an individual exhibits when signing his or her name (or other phrase).
Dynamic signature devices should not be confused with off-line electronic signature capture systems that are used to capture a graphic image of the signature and are common in locations where merchants are capturing signatures for transaction authorizations.
Data such as the dynamically captured direction, stroke, pressure, and shape of an individual’s signature can enable handwriting to be a reliable indicator of an individual’s identity (i.e., measurements of the captured data, when compared to those of matching samples, are a reliable biometric for writer identification.)
The first signature recognition system was developed in 1965. Dynamic signature recognition research continued in the 1970s focusing on the use of static or geometric characteristics (what the signature looks like) rather than dynamic characteristics (how the signature was made). Interest in dynamic characteristics surged with the availability of better acquisition systems accomplished through the use of touch sensitive technologies.
In 1977, a patent was awarded for a “personal identification apparatus” that was able to acquire dynamic pressure information.
Dynamic signature recognition uses multiple characteristics in the analysis of an individual’s handwriting. These characteristics vary in use and importance from vendor to vendor and are collected using contact sensitive technologies, such as PDAs or digitizing tablets, which acquires the signature in real time.
Most of the features used are dynamic characteristics rather than static and geometric characteristics, although some vendors also include these characteristics in their analyses. Common dynamic characteristics include the velocity, acceleration, timing, pressure, and direction of the signature strokes, all analyzed in the X, Y, and Z directions.
The X and Y position are used to show the changes in velocity in the respective directions while the Z direction is used to indicate changes in pressure with respect to time.
Some dynamic signature recognition algorithms incorporate a learning function to account for the natural changes or drifts that occur in an individual’s signature over time. The most popular pattern recognition techniques applied for signature recognition are dynamic time warping, hidden Markov models and vector quantization. Combinations of different techniques also exist.
The characteristics used for dynamic signature recognition are almost impossible to replicate. Unlike a graphical image of the signature, which can be replicated by a trained human forger, a computer manipulation, or a photocopy, dynamic characteristics are complex and unique to the handwriting style of the individual. Despite this major strength of dynamic signature recognition, the characteristics historically have a large intra-class variability (meaning that an individual’s own signature may vary from collection to collection), often making dynamic signature recognition difficult. Recent research has reported that static writing samples can be successfully analyzed to overcome this issue.