Abstarct: Face recognition is one of the most active research fields of machine vision and pattern recognition, which is widely used in many fields such as identification, access control, and public security. The development of deep learning techniques, access to large-scale face datasets and production of high-power processing systems, the performance of face recognition systems has improved significantly. However, face recognition systems still do not perform very satisfactorily when facing challenges such as pose variety, different illumination, low resolution, and occlusion. Image occlusion is one of the most challenging face recognition problems. In this challenge, the appearance of the face changes significantly and the identity features of the face are lost, which makes it difficult to recognize the face. Using a mask blocks a large part of the face, including the nose and chin. Hence, it is considered the most difficult challenge of facial occlusion. One of the methods of face detection in case of obstruction is face reconstruction and restoration. In recent years, GAN-baxsed networks have performed very well in the field of image restoration and reconstruction. In the proposed method of this thesis, first the mask area is detected and then this part is reconstructed and the face without mask is created. Then the reconstructed face is given to the face recognition system. The proposed deep network architecture in face reconstruction is baxsed on GAN. The output of the network, in addition to producing a high-quality image, preserves the identity features of the area under the mask, that is, the nose and mouth. Therefore, the proposed method increases the accuracy by about 30% compared to the masked image and by about 8% compared to the compared methods. Also, the quantitative criteria of SSIM, PSNR and FID indicate the proper performance of the proposed method in the mask area restoration.