In forensic investigations, the restoration of obscured facial images plays a crucial role in identifying suspects involved in criminal activities. The purpose of inpainting an image is to replace the missing areas of a damaged image with realistically created material. Early method for image inpainting involves the use of the traditional methods which are patch-based and diffusion-based methods. When the missing region is vast or has complicated structures, they frequently create semantically incoherent content. The development of deep learning algorithms has revolutionized the field of face restoration, allowing for more accurate and reliable reconstruction of occluded face employing deep learning methods such as Generative Adversarial Networks (GANs) can be dealt with for large occluded regions. We propose a Residual Network for image inpainting. The two neural networks that make up our Generative Adversarial Networks are the discriminator and the generator. The generator is made up of an encoder and decoder design with a skip connection between them. And the discriminator is a global discriminator. The model was trained and testing on CelebA dataset. The PSNR and SSIM are used to assess the model’s performance.