Score level fusion of classifiers in off-line signature verification

Offline signature verification is a task that benefits from matching both the global shape and local details; as such, it is particularly suitable to a fusion approach. We present a system that uses a score-level fusion of complementary classifiers that use different local features (histogram of oriented gradients, local binary patterns and scale invariant feature transform descriptors), where each classifier uses a feature-level fusion to represent local features at coarse-to-fine levels. For classifiers, two different approaches are investigated, namely global and user-dependent classifiers. User-dependent classifiers are trained separately for each user, to learn to differentiate that user’s genuine signatures from other signatures; while a single global classifier is trained with difference vectors of query and reference signatures of all users in the training set, to learn the importance of different types of dissimilarities. The fusion of all classifiers achieves a state-of-the-art performance with 6.97% equal error rate in skilled forgery tests using the public GPDS-160 signature database. The proposed system does not require skilled forgeries of the enrolling user, which is essential for real life applications