Face Recognition has been extensively embedded in numerous systems in our daily lives and has attained significant performance for high-resolution images in current years. However, till now recognizing faces captured by surveillance cameras is a challenging task due to low and very low resolutions. To address this issue, this paper has presented two Convolutional Neural Network (CNN) based face recognition systems and a Vision Transformer (ViT) based face recognition system. The proposed systems are capable of processing and encoding of low as well as very low-resolution images in such a way that they can be fit for the automated recognition process. To validate proposed systems experiments have been conducted on face images of varying resolutions ranging from very low to low (10 × 10, 20 × 20, 30 × 30, and 40 × 40) of the UMDAA-02 dataset. The proposed CNN model achieved 51.96% recognition accuracy for very low resolution images (10 × 10) while the proposed ViT model achieved 95.73% recognition accuracy for the same image resolution. The proposed CNN model achieved 98.65% accuracy for low resolution images (40 × 40) while the proposed ViT model achieved 98.22% accuracy for the same image resolution. The experimental outcomes established that ViT is more robust compared to CNN towards very low-resolution face recognition. Also, this finding suggests that ViT has the potential to be an effective tool in security settings for accurate person recognition paving the way for better safety and security.