Comparativo de algoritmos clássicos de aprendizado de máquina em um problema de reconhecimento de faces

Abstract

With the increased availability of face data, machine learning becomes a very relevant tool for identify people in various applications. This work aims to evaluate the use of traditional machine learning algorithms on features extracted from face images, in order to perform their recognition. The extracted features are based on the Principal Component Analysis (PCA), that represent each face as a n-dimensional vector, where n is less than the image size. The evaluated algorithms are distance classifiers, decision trees, support vector machines and neural networks, which were evaluated in their different parametrizations. The best accuracy results were obtained by the support vector machine and artificial neural networks models (99%), with the latter presenting the highest accuracy (97%) when the same model was tested in a new database.