Abstract:

A new curve similarity calculation model is presented for online signature verification. Considering online signature trajectory data as a plane curve, a curve similarity model is built based on the similarity definition, similarity transformation and curve similarity distance. The comparison signature curve is resampled according to the effective point number of the reference one. In the calculation of curve similarity, the sectional matching is applied. In the range of each comparison curve interval to match, the sectional curve similarity distance is calculated by the genetic algorithm. The optimal sectional matching is found in the corresponding section of the reference curve. Then the sectional similarity distance is obtained and the estimate sectional similarity score is calculated. The mean of all similarity scores is defined as the curve similarity measure. The proposed algorithm is tested, and the equal error rates are 10.92% and 2.89% for SVC2004 Task1 and SUSIG Blind databases, respectively.