Abstarct: Face detail restoration from degraded images remains one of the fundamental challenges in the field of computer vision, with significant implications for applications such as surveillance, digital forensics, and multimedia systems. Although conventional deep learning-baxsed approaches have achieved promising results in image restoration, they often suffer from limitations including inadequate recovery of complex facial characteristics, insufficient generalization to different types of image degradation, and difficulties in preserving both perceptual quality and identity-specific facial features. To address these challenges, this study proposes an enhanced face restoration frxamework that leverages advanced generative adversarial architectures, namely GFP-GAN and ESR-GAN, to synergistically exploit the strengths of both models in feature extraction and perceptual quality preservation. In the proposed approach, perceptual loss is combined with conventional reconstruction metrics to achieve a balance between structural fidelity and perceptual authenticity. This dual-objective strategy enables the model not only to reconstruct fine pixel-level details but also to recover high-level semantic facial attributes that are essential for identity preservation and facial exxpression consistency. For training and evaluation, the CelebA dataset, one of the most widely used facial image datasets, was employed. In addition, to assess the model's performance under real-world conditions, a supplementary collection of facial images from volunteer subjects was gathered and utilized. To evaluate robustness, a dataset containing systematically degraded facial images was generated. The degradation protocol included the addition of Gaussian noise to simulate sensor imperfections and compression artifacts, Gaussian blur to emulate motion blur and defocus effects, and low-light conditions created through gamma correction and contrast reduction. This comprehensive degradation frxamework enabled a realistic and thorough evaluation of the proposed model under conditions commonly encountered in practical applications. Experimental results demonstrate the superiority of the proposed frxamework across several established quantitative and qualitative evaluation metrics. Significant improvements in Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index Measure (SSIM) indicate enhanced pixel-level reconstruction accuracy and structural preservation. Furthermore, improved LPIPS scores confirm superior perceptual quality and greater visual realism. Statistical analyses conducted on a wide range of test samples further validate the robustness and generalization capability of the proposed method under various degradation scenarios, including low-light environments and heavily corrupted images. These findings highlight the effectiveness and practical applicability of the proposed approach for real-world face restoration tasks.