|
« Back
Signature Verification
As one of the biometric authentication methods, signature has been widely accepted in our real life, because it is user-friendlier than fingerprint, iris, retina and face. On-line signatures are acquired using digitizing tablet that captures both temporal and spatial information, such as coordinates, pressure, inclinations, etc. It is not possible to reproduce the online features even if a person's handwriting is copied from an already existing document. Therefore it is significantly more difficult to circumvent online handwriting and signature verification systems. However, two aspects pose challenges in the field of online signature verification. Firstly, intra-personal variation can be large. Our recent study about the distinctiveness of signature features indicate that the speed, pressure and inclinations pertaining to the signatures made by the same person can differ greatly making it quite challenging to extract consistent features. Secondly, we can only expect few samples from one person and no forgeries in practice. This makes it difficult to determine the consistency of extracted features.
Existing matching that use Euclidean distance, DTW (Dynamic Time Warping) or other distances measure provides scores that have relative meaning. That is, the distance itself cannot give us any information about similarity without comparing it with other distances. We have developed a new metric that is intuitive and directly correlates with the similarity of the signatures. We extend R2 measure used in SLR (Simple Linear Regression) of 1D signals to ER2 for multidimensional sequence matching. Also, we incorporate the optimal alignment by DTW (Dynamic Time Warping) into ER2 to make it robust on signature verification. We have tested the algorithm on random forgeries, resulting in 0.2% equal error rate (EER) and 0.8% EER when using global and user dependent thresholds respectively. Experimental results show that ER2 coupled with optimal alignment outperforms DTW-based curve matching on online signature verification.
Publications:
- H. Lei and V. Govindaraju, "A Comparative Study on the Consistency of Features in On-Line Signature Verification", SSPR 2004.
- H. Lei and V. Govindaraju, "A Comparative Study on the Consistency of Features in On-Line Signature Verification", SSPR 2004.
- H. Lei, S. Palla, V. Govindaraju. ER-squared: an Intuitive Similarity Measure for On-line Signature Verification. The 9th International Workshop on Frontiers in Handwriting Recognition (IWFHR'04), Tokyo, Japan, Oct., 2004
|
|
 Human iris patterns have a high degree of randomness and uniqueness.
|
