Combination of OC-LBP and Longest Run Features for Off-Line Signature Verification

In this paper, we propose new data features to improve the off-line handwritten signature verification. The proposed features combine advantages of LBP and topological characteristics. Specifically, the Orthogonal Combination of LBP, which provides an LBP histogram with a reduced size, is combined with a topological descriptor that is called longest run features. The verification task is achieved by SVM classifiers and the performance assessment is conducted comparatively to the basic LBP descriptors. Results obtained on both GPDS 300 and CEDAR datasets show that the proposed features improve the verification accuracy while reducing the data size.