Signatures have become ubiquitous as means of identification and are used for authorizing many transactions. Usually, handwritten signatures are verified by careful visual observation. Automating this process will reduce the amount of time taken for verification and may also improve the efficiency.
Signature Verification techniques can be classified into two types - online and offline. In online signature verification, a special instrument is used to record "dynamic" information of the signature such as pressure applied, velocity, etc. This "dynamic" information is then used to authenticate the signature. In offline signature verification, a scanned image is used to authenticate the signature. This problem, in general, is more difficult, since important "dynamic" information is not available, but is attractive due to the cost benefits.
Many techniques exist in literature for offline signature verification, but almost all of them perform poorly when confronted with skilled forgeries. We are developing a technique that will be accurate in detecting skilled forgeries as well as simple and casual forgeries. Preliminary results obtained are promising.
The system currently uses a modified Dynamic Time Warping (DTW) algorithm to match certain features extracted from the signature. A cost of matching is computed and then a hard decision threshold is used to verify the claim.
A signature database comprising of about 1000 signatures collected from 100 people was used for testing the system. 10 samples were collected from each person and were used for training the system.
Results:
A verification system is characterized by the False Rejection Rate (FRR) and False Acceptance Rate (FAR). Both the FRR and the FAR are a function of the threshold value used.
The system was found to be 98% accurate in identifying genuine signatures and casual forgeries. The system functioned poorly in resolving skilled forgeries (65% accuracy).
A GUI has also been developed to facilitate the use of this signature verification system. A snapshot of the screen presented to the user is shown below:
Genuine Signature:
Simple Forgery:
