Role of sub-trajectories in online signature verification

In this paper, we have provided a partitioned based technique which can increase the efficiency of an existing technique in which each partition is called as a sub-trajectory. To implement it, eighty features are extracted from signature trajectories and categorized into four feature sets as static, kinematics, structural and statistical. We have used these four feature categories and their possible combinations on two different algorithms. An important outcome is observed as the EER decreases with increase in sub-trajectories to an optimum level and behaves in reverse direction afterwards which suggests that for any matching algorithm, the present technique can further reduce the error rate to an optimum level. These experiments are performed on the benchmark database SVC 2004 TASK 2 which contains forty genuine signatures of each forty writers and forty skilled forgeries from five different writers. The experiments are discussed in detail for change in the EER with change in each subsequent sub-trajectory level for all feature sets and the results prove that the technique using sub-trajectories improving the EER by a significant average amount of 1.18 with increase in one sub-trajectory level for all the eighty features and 1.5 for the features of categories kinematics and structural when taken together.