Skip to Main Content
In this work, a novel segmentation scheme for online handwritten signatures is presented in order to extract local characteristics that are consistent among the different signature samples. In the new segmentation scheme, the segmentation points of the reference/master signature are allocated at the minimum velocity points or ldquovalleysrdquo. To avoid an excessive number of segmentation points, a constraint factor of minimum segment size is applied while extracting the valleys. Then, the correspondent segmentation points in each of the test or other enrollment signatures are allocated by finding the minimum velocity points occurring within a window of points centered at each segment-boundary-point of the master signature. The paper also presents an algorithm to select the master/reference signature sample among the enrollment signature samples for each writer. According to the selection criteria, the master/reference signature is the signature that contains the most repeated valleys locations. The evaluation of the new segmentation method showed significant improvement over a benchmark method, developed in earlier work, whose segmentation was based on the dynamic time warping technique.