Feature selection method for offline signature verification

Abstract

Signature verification is defined as one of the biometric identification method using a person’s signature characteristics. The task of verifying the genuineness of a person signature is a complex problem due to the inconsistencies in the person signatures such as slant, strokes, alignment, etc. Too many features may decrease the False Rejection Rate (FRR) but also increases the False Acceptance Rate (FAR). A low value of FAR and FRR are required to obtain accurate verification result. There is a need to select the best features set of the signatures attributes among them. A combination of the current global features with four new features will be proposed such as horizontal distance, vertical distance, hypotenuse distance and angle. However, the value of FAR may increase if too many features are used which result a slow verification performance. In order to select the best features, the difference between the mean of the standard deviation ratio of each feature will be used. The main objective is to increase the accuracy of verification rate. This can be determined using best features set selected during the features selection process. A selection of signature set with strong feature sets will be used as a control parameter. The parameter is then used to validate the results.