##
###
DWT Domain On-Line Signature Verification
[chapter]

Isao Nakanishi, Shouta Koike, Yoshio Itoh, Shigang Li

2011
*
Biometrics
*

verification performance tends to be degraded since the on-line signature is a dynamic trait. We have proposed a new on-line signature verification method in which a pen-position parameter is decomposed into sub-band signals using the discrete wavelet transform (DWT) and total decision is done by fusing verification results in sub-bands (Nakanishi et al., 2003; . The reason why we use only the pen-position parameter is that detecting functions of other parameters such as pen-pressure,
## more »

... de, and/or pen-direction are not equipped in the PDA. However, since the signature shape is visible, it is relatively easy to forge the pen-position parameter by tracing genuine signatures by others. In the proposed method, individual features of a signature are enhanced and extracted in the sub-band signals, so that such well-forged signatures can be distinguished from genuine ones. Additionally, in the verification process of the proposed method, dynamic programming (DP) matching is adopted to make it possible to verify two data series with different number of sampled points. The purpose of the DP matching is to find the best combination between such two data series. Concretely, a DP distance is calculated in every possible combination of the two data series and as a result the combination which has the smallest DP distance is regarded as the best. But there are problems in use of the DP matching. The DP distance is obtained as dissimilarity; therefore, signatures with large DP distances are rejected even if they are of genuine. For instance, in a pen-tablet system, a pen-up while writing causes large differences in coordinate values of pen-position and so increases false rejection. On the other hand, signatures with small DP distances are accepted even if they are forgery. The DP matching forces to match two signatures even if either is forgery. It increases false acceptance. Consequently, we propose simply-partitioned DP matching. Two data series compared are divided into several partitions and the DP distance is calculated every partition. The DP distance is initialized at the start of a next partition, so that it reduces excessively large DP distances, that is, the false rejection. On the other hand, limitation of combination in matching is effective for rejecting forgeries; therefore, it reduces the false acceptance. There is another important problem when we use the DP matching. The DP distance is proportional to the number of signature's sampled data, that is, signature complexity (shape), so that if it is used as a criterion in verification, each signature (user) has a different optimal threshold. But, it is general to use a single threshold commonly in an authentication system. If the common threshold is used for all signatures, it results in degradation of verification performance. Therefore, we have studied threshold equalization in the on-line signature verification (Nakanishi et al., 2008) . We propose new equalizing methods based on linear and nonlinear approximation between the number of sampled data and optimal thresholds.

doi:10.5772/17486
fatcat:3rmchhro5rc4nopvqbqakxztqy