Improving Face Recognition Rate Based on Histogram of Oriented Gradients and Difference of Gaussian

Face recognition is a widely used identification method in the machine learning field because face biometrics are distinctive enough for detection and have more accessibility compared to other biometrics. Despite their merits, face biometrics have various challenges. Mainly, these challenges are divided into local and global categories. Local challenges can be addressed using sustainable methods against change while global challenges such as illumination challenges require powerful pre-processing methods. Therefore, in this study, a sustainable method against light changes has been proposed. In this method, two stages of the Difference of Gaussian have been utilized for the illumination normalization. Then, the features of the normalized image are extracted using Histogram of Oriented Gradient (HOG) and the feature vectors are classified using 3 k-nearest neighbor classifiers and the support vector machine with linear kernel, and the support vector machine with Radial Basis Function (RBF) kernel. Testing the proposed method on Computer Vision and Biometric Laboratory (CVBL) data indicated that the recognition rate, at best for the illumination challenge in the whole face and a part of the face is 98.6 % and 97.9% respectively.