APPLYING MODERN CNN ARCHITECTURES TO BUILD AN EFFECTIVE FACE DETECTION MODEL

Abstract

Facial recognition is seen as one of the big problems in computer vision. This field has received much attention due to the progress in artificial intelligence and deep learning. The availability of diverse data sets makes training of the models more accurate and as a result training of these models is important for the development of facial recognition systems in many industries as security and healthcare. The goal of this research is to develop an intelligent facial recognition system based on the pre-trained models DenseNet201, Darknet-19, InceptioResNetV2, and MobileNetV2, which are then fine-tuned in a supervised manner on a custom dataset in order to improve their utility. The research also adds to the body of knowledge on the accuracy and efficiency of these models for deployment in environments where speed and accuracy are paramount. Experimentation showed the DenseNet201 model to outperform the other models in terms of overall performance.

The model achieved exceptional results, attaining 92.75% accuracy, 86.09% precision, and a true positive rate (TPR) of 85.5%. Such outcomes demonstrate the capability of the DenseNet201 model and justify its selection for this research classification task, as it is trustworthy in feature and pattern extraction.