Deep Learning based Face Recognition (FR) methods have developed rapidly with great achievements, but they usually require plenty of high-quality data to build the model. When the dataset is small, classical FR techniques, such as Eigenface, are preferable. Therefore, we aim to explore the most representative feature of human faces and the best-matching classifier when there is only one training face for each face subject in the database. We proposed a Histograms of Oriented Gradient (HOG) and K Nearest Neighbour (KNN) combined approach, which is tested to be the most accurate method in this environment.
Use Evaluation.m to start.