Face Recognition Using Pre-Trained CNN Embeddings and a Custom Deep Neural Network

Facial recognition technology has seen rapid advancements and development, especially with the adoption of deep learning algorithms. In this method, a custom Deep Neural Network (DNN) model is trained on two different datasets. This study also focuses on a video-based facial recognition system to efficiently and accurately recognize any specific person who is in the dataset. There is also shown the comparison with Convolutional Neural Networks (CNNs), along with traditional methods like Eigenface, Fisherface, and Local Binary Patterns Histogram (LBPH), as well as other deep learning approaches like Support Vector Machines (SVM) and Multilayer Perceptron (MLP). This study achieved state-of-the-art accuracy of 94 % on a larger dataset with 5,817 labels from online and 99.37 % on a self-collected smaller dataset with 68 labels and 1,564 augmented samples. These findings highlight the custom DNN model's suitability for real-world facial recognition applications, offering high accuracy and robustness across different datasets.