Handwritten Signature Forgery Detection Using PCA and Boruta Feature Selection

Despite the development of identity detection using biometrics in the field of financial transactions, the handwritten signature remains the most commonly used to this day. The main challenge is that each person's signature may be distinctive, on the other hand, many difficulties aroused because two signatures created by the same individual may appear to be extremely identical. This similarity allows the imposters to claim a forged identity. In this paper, an off-line handwritten forgery detection method is introduced using traditional machine learning rather than deep learning methods to fulfill the need for a simpler model for saving both computation time and computation resources. The proposed method uses Histogram of Gradients (HOG) as a feature extraction method and Principal Component Analysis (PCA) to reduce the large extracted features number and Support Vector Machine (SVM) as a classifier. Another approach has been used by using Boruta feature selection for further reduction of feature numbers. CEDAR dataset has been used in this paper and the results were 99.24 % and 98.79 % in terms of accuracy for the two proposed methods respectively.