Offline signature verification using geometric and orientation features with multiple experts fusion

This paper reports a weighted fusion of multiple classifiers for offline signature verification using geometric and orientation features. The proposed system uses three different classifiers for identity verification, namely, Gaussian empirical rule, Mahalanobis and Euclidean distance metrics. Initially, Geometric global and local features are extracted from signature image. Further, a novel feature extraction technique is applied to signature image for extraction of orientation features. These feature sets are then fused and make a concatenated feature set which is then passed through the three classifiers. Matching scores obtained from these three classifiers are finally fused using weighted sum rule. The proposed system is tested on IIT Kanpur signature database which consists of 1800 offline signatures. The experimental results are found to be convincing and encouraging. The aim of the proposed system is to provide such a system which can overcome the problem of skilled forgery detection efficiently with less computational complexity.