Science & Enterprise subscription

Please share Science & Enterprise

RSS
Follow by Email
Facebook
Facebook
Google+
Twitter
Visit Us
LinkedIn

Affiliations

Smart Watch Software Designed to Verify Signatures

Smart watch

(Oliur Rahman, Pexels)

30 January 2017. Engineering researchers in Israel created software that enables smart watches and fitness bands to capture wrist movements for verifying a person’s signature. A team from Tel Aviv University and Ben-Gurion University of the Negev recently published results of a study testing the software, in the research repository arXiv, and filed a patent on the technology.

The researchers led by Tel Aviv industrial engineering professor Erez Shmueli are seeking a more feasible method for verification of an individual’s signature that takes advantage of the growing number of devices worn on the wrist. Despite advances in biometrics, a person’s handwritten signature is still the most widely used method of identity verification in business transactions. The continuing use of paper checks by large numbers of people, in particular, still offer significant opportunities for fraud. The authors cite statistics from American Bankers Association that in 2011 paper checks accounted for $34 trillion in transactions, with losses due to check fraud amounting to $1.2 billion.

Signature verification is done either offline, with visual comparison to a physical signature sample on file, or online with the signature entered on a digitized surface using an electronic pen or stylus. The online approach offers advantages in speed and efficiency, but often requires a separate digitizer, such as a tablet, to capture the signature image for transmission.

Shmueli and colleagues believe the handwritten signature’s uniqueness to a person and the increasing use of smart watches present an opportunity to create a simple identity verification process that’s also easy to implement. The authors hypothesize that a person’s signature represents a pattern of muscle movements developed over a period of years and is difficult to imitate. In addition, those movements can be captured with a device worn on the hand or wrist.

Smart watches and fitness trackers, say the researchers, could very well be those devices. The authors cite data from Global Web Index showing in 2014, 1 in 6 Internet users worldwide wear a smart watch or fitness band, with that number continuing to grow. One estimate puts the global market of smart watches in 2020 at 373 million devices.

The team took the approach of capturing the dynamics of writing a signature — the hand and wrist movements producing the signature — as the identification mechanism. This approach makes it possible to apply the techniques both to pen-and-ink signatures on paper documents as well as electronic signatures captured on a separate tablet. The researchers applied machine learning to both genuine and forged signatures, enabling identification of movement patterns and properties of real and bogus signatures, to classify the signatures accordingly. In other words, the system identifies the characteristic movements of writing a genuine signature, compared to movements indicating a forged signature, rather than comparing movements for making the signatures themselves.

To build its knowledge base for training the software, the team asked 66 volunteers at Tel Aviv University to write samples of 30 signatures on a Samsung electronic tablet. The first 15 samples were their own signatures, while the next 15 signatures were forgeries. Participants were asked not only to forge the signature, they were shown videos of the real signatures being written, with volunteers given bonus points for making particularly good forgeries. The volunteers were given plenty of time, including replays of the videos, and were asked to indicate what they considered their best forgeries.

And while all of these signatures were being written, participants wore a Microsoft fitness band, with accelerometers capturing the hand and wrist movements. The researchers ended up with a total of 1,980 samples to train the software, from which they applied machine learning algorithms. The authors sampled 5 of each participant’s real signatures for reference, then tested the software’s ability to distinguish between movements producing real signatures to those making forged signatures.

The team used statistical tests indicating the software’s (1) ability to distinguish between real and forged signatures, and (2) return erroneous results. The findings show overall the software could distinguish between real and bogus signatures, even skilled forgeries, with a high degree of confidence, returning a score of 0.98 where 1.00 is perfect discrimination. The false result score was 0.05, where 0.00 represents no erroneous results.

Their next research steps, says Shmueli in a Tel Aviv University statement is “to compare our approach with existing state-of-the-art methods for offline and online signature verification.” The team applied for a patent on the technology, to put the software on the road to commercialization.

Read more:

*     *     *

Please share Science & Enterprise ...

Comments are closed.