Pre-Trained Model Based Face Classification

The one of the most important applications of computer vision and pattern recognition technologies, facial recognition has made significant progress in recent times. These applications have resonated in many areas. For this purpose, various methodologies have been developed and compared based on various factors and aspects to find the most suitable methodology. Facial recognition technology has progressed rapidly, from traditional methods to deep learning applications. However, most of these methods are still limited by variations in constraints such as lighting, expressions, orientation, and other practical complexities. The need to examine all these methods and evaluate them critically and comprehensively has emerged. The purpose of this article is to examine the different but comparable features of deep learning methodologies applied to facial recognition systems, and to examine, evaluate, and explain their strengths and weaknesses. To this end, images taken from a small scale data set of 10 people were classified using pre-trained networks. The classification process, performed using four different pre-trained network models, achieved a performance of 98%. It has shown that it can be used for low profile systems such as Raspberry Pi Zero.