Enhanced Facial Recognition Using CNN and GoogleNet: A High-Performance Framework

Facial recognition technology has become crucial in enhancing security, authentication, and interactions between people and devices. This paper introduces an Advanced Facial Recognition System that combines Convolutional Neural Networks (CNNs) with GoogleNet to improve classification performance. The suggested model was assessed using the CK+ (Cohn-Kanade) dataset, which is a commonly referenced benchmark for facial recognition and emotion evaluation. The machine learning model is evaluated using different parameters. The Training Accuracy, Validation Accuracy, and Testing Accuracy are around 99%, 100%, and 89.70%. The Precision, Recall, Specificity, and F1 Score are achieved in the range of 91.00%, 88.50%, 92.20%, and 89.70%. The results indicate a very good balance between the positive and negative false values. The model improved using SGDM and a Cross-Entropy Loss function, and was able to differentiate different facial identities very well, as proved with the help of a 0.94 ROC-AUC score. The training time required to execute the simulation is around 45 minutes. The batch sizes are kept at 32 and an LR of 0.01. The model attains convergence inside 6 epochs. The proposed model is compared with many machine learning models, such as ResNet-50, VGG-16, and MobileNet, on the basis of testing accuracy, precision, and specificity. The given model maintains lower computational costs and achieves faster convergence. The testing accuracy of ResNet-50 is 87.50% but it needs large computational resources. The VGG-16 model has an accuracy of 86.20% but it has an overfitting problem. The testing accuracy of Mobile Net is 85.10%, and it is used for mobile applications. Its accuracy is low compared to the CNN-GoogleNet. The findings confirm the proposed model, which is hybrid, combining CNN and GoogleNet, strikes a balance between precision and efficiency, making it well-suited for real-time facial recognition.