Techniques for static handwriting trajectory recovery: a survey
DAS '10 Proceedings of the 9th IAPR International Workshop on Document Analysis Systems
| Hi-index | 0.00 |
Off-line signature verification is an important issue in establishing the authenticity of bank cheques and other official documents. However, because of the loss of dynamic and temporal information in off-line signatures, the performance of off-line signature verification systems is always not as good as that of on-line ones. The main objective of this thesis is to investigate the possibility of recovering the dynamic information from off-line signatures and the effectiveness of applying the recovered dynamic information in off-line signature verification. In fact, there are various kinds of dynamic information in an on-line signature, such as stroke sequence, velocity and acceleration. But the relatively important one is the stroke sequence, as none of the other dynamic information can be recovered without the stroke sequence. This research work is focused on the recovery of the stroke sequence. In this thesis, a universal writing model (UWM) is developed so as to model the writing sequence of people. The model considers the probabilistic relationship of the writing strokes, such as the probability of the writing direction of a stroke, the probability of being the first writing stroke, and the probability of being the next writing stroke. The probabilities are computed in the training stage from a set of on-line writing signatures. By using the estimated probabilities, the UWM can recover the sequence of the extracted strokes from an off-line signature, and the recovered sequence is close to the writing style of the common people. The recovered sequence is only an approximation. It cannot guarantee that the recovered sequence is identical to the originally signed sequence. The question of how similar the recovered sequence compared with the original one is raised. However, there is no existing methods for making such comparison. Although ranking analysis may be helpful, the distances defined are not convincing in the evaluation. In this thesis, a directed connection measurement is then proposed to compare the recovered sequences, or even evaluate the performance of stroke recovery algorithms in general. The proposed measurement includes the measurements in terms of both direction and connection of the sequence. The direction measurement is based on the consecutive arrangements of the items, while the connection measurement is developed based on both the number of disconnections and the Feigin and Cohen model (FCM) in ranking analysis. Finally, this thesis investigates the use of recovered stroke sequences in off-line signature verification. (Abstract shortened by UMI.)