Offline signature recognition is one of the most difficult methods of distinguishing patterns, as it requires extrapolation of these skilled processes of forgery that are not available during training, and its challenges also include a few samples and significant variation in the sample category. The use of handwritten signatures occupies multiple areas of everyday life and is largely a secure means of personal identification.
In this research proposed method has been put forward to distinguish signatures whether they are forged or not by using gray level co-occurrence matrix (GLCM) for extract texture of each signature sample then extract features of these samples using Haralick features (Energy, Entropy, Contrast, and Homogeneity). The next step is classification step, at this step used extracted features to specify the signature is forged or not, then use convolution neural network (CNN) as classifier to specify the class of signature.
the result in this research the system works more efficiently with signatures (UTSig) which contains 115 persons for each person 27 sample of signature, now 70% of them use to training and the rest use to testing. After testing the accuracy of the system is98% for signature verification (forgery or not), and 99% for recognize class of signature.