An Efficient Offline Signature Verification Method Based on Improved Feature Extraction

As a kind of biometric system, handwritten signature verification has always been a research hotspot. Offline signature verification (OSV), although heavily researched in the past, remains shortcomings such as incomplete feature extraction, too many training samples and inconvenient promotion in practical applications, especially in distinguishing genuine signatures from skilled forgeries, which may be very close in appearance to the genuine ones. To solve these problems, in this paper, we propose a writer-dependent (WD) offline signature verification system, which extracts features from signature images through Gray Level Co-occurrence Matrix (GLCM) and Improved Local Binary Pattern (ILBP) respectively, and these features are fused with geometric features, and finally we use Support Vector Machine (SVM) for classification. The MCYT75 and CEDAR public datasets are used to validate the proposed system, and the experimental results show advantages in both accuracy and average error rate (AER) compared with other existing methods. The accuracy rate (AR) reaches 97.58% and 97.0%, respectively, and 5.64% AER and 3.49% AER, respectively. The experimental results compare favorably with many of the state-of-the-art in handwritten signature verification, proving the usefulness of our designed system.