Du Visage Vers Une Image De Qualite Naturelle

Abstract

From the face to a natural quality image Summary This master’s thesis focuses on the development of a face image super-resolution system aimed at reconstructing low-resolution images into high-resolution natural quality images. Using deep learning techniques, particularly convolutional neural networks (CNN), our research has led to a high-performing system. Our study may have significant implications, particularly in the fields of facial recognition, video surveillance, high-definition television, and security systems. Re- constructed facial images can improve the accuracy of facial recognition systems and offer a better visual experience. However, the computational time and power required for real-time super-resolution are often insufficient for facial images to reproduce. The main contribution lies in improving existing techniques and designing a sys- tem that can obtain high-resolution facial images with increased accuracy and fidelity. Overall, the results of our study have shown a strong improvement in terms of visual quality, edge sharpness, and preservation of facial details. We were able to surpass existing methods by designing a specific CNN for face image super-resolution. The future research directions could therefore focus on improving these aspects and exploring new network architectures. In summary, this research has made a significant contribution to the development of a super-resolution system based on CNN for facial images. The results obtained are interesting and give hope for improvements in various applications requiring high- resolution images. However, further research is needed to overcome the identified limitations and continue exploring the possibilities of image super-resolution.