Masked face recognition has become crucial for secure authentication, especially after the COVID-19 pandemic. Traditional face recognition systems struggle with obstacles like masks, which lower accuracy. This study introduces a lightweight CNN model designed for real-time mobile use, utilizing MobileNet and TensorFlow Lite. The model is trained and tested on the MadFaRe, RMFD, and SMFRD datasets to improve reliability and performance, incorporating data augmentation techniques to enhance robustness under different lighting conditions and occlusions. Our approach achieves 90.4% accuracy at 45 FPS, ensuring both efficiency and reliability. While more complex architectures like Vision Transformers (ViTs) and EfficientNet offer high accuracy, they are computationally expensive for mobile devices. Future work will explore EfficientNet-based feature extraction to further optimize performance. Additionally, ethical concerns related to privacy and bias in facial recognition systems are considered, ensuring responsible AI deployment. Our model offers a practical, accurate, and efficient solution for real-world applications.