Optimizing Face Recognition Accuracy with HOG Features and SVM Classifier: A Study on ORL and Yale Databases

Face recognition is a technology that identifies individuals by analyzing and comparing facial features, this technology relies on advanced algorithms and machine learning to improve accuracy and efficiency. This paper proposes a face recognition using a histogram of orientated gradients (HOG) features and a Support Vector Machine (SVM) classifier, testing on ORL and Yale databases. At first to evaluate the (SVM) model we used different train percentages 40%, 60%, and 80%, also to ensure a soft sliding window of HOG features parameters we resized the images in both databases into (128 X 64 pixels). Secondly, a histogram of orientated gradients (HOG) was performed to extract features, since it is robust to variations in lighting, pose, and facial expressions. Finally, a Support Vector Machine (SVM) is used to classify the extracted feature vector, since it is suitable for small datasets. The proposed method has achieved a recognition rate of 96.25% and 99.98% on ORL and Yale databases, respectively at 80% train percentage, using cell size - (10x10 pixels) and block size - ($4 \times 4$ cells) of the HOG parameters with the radial basis function (RBF) kernel of the support vector machine (SVM) classifier, which proved encouraging results, in comparison with the default parameters, and other related works.