DYNAMIC POSITIONAL WARPING: DYNAMIC TIME WARPING FOR ONLINE HANDWRITING

摘要

This paper addresses the problem of dynamic time warping (DTW) causing unintended matching correspondences when it is employed for online two-dimensional (2D) hand-writing signals, and proposes the concept of dynamic positional warping (DPW) in conjunction with DTW for online handwriting matching problems. The proposed DPW allows subsignal translations without any additional costs when their starting points are matched to other points through the matching process. Because the movement of subsignals is cost-free, except for the distance between the two starting points, an adequate movement - finding and matching similar subsignals - could significantly reduce the matching cost. This feature causes a tendency to match with the least amount of subsignal movement in the result. In order to evaluate the proposed method of solving this problem, two experiments were conducted: an accuracy test for matching similar handwriting and a utility test for signature verification. For the former, we proposed the new concept of ideal matching between two handwriting signals using the affine transformation; the results of this test verified the superiority of the proposed DPW over conventional methods in reducing matching errors. Further, DPW significantly outperformed the conventional methods in the latter test.