This paper presents the development of an algorithm focused on face completion implemented by means of two techniques: the first one is semantic segmentation, responsible of the detection of the parts of the face, for which it is proposed to use a SegNet, where in its training and subsequent fine-tuning, an accuracy of 97.51% in the segmentation of the categories of interest is reached (mouth, eyes, face and background). The second one is represented by the geometric features, whose function is the location of the missing parts of the face, in which 3 cases are established, depending on which part of the face is removed, so different relationships of the characteristics of the face are set to strengthen the development in terms of variable face sizes. With the union of these two techniques, it is possible to implement an algorithm that is sufficiently robust to perform face completion, achieving successful results in reconstructed faces, even with a low sensitivity to tilt, rotation and face size.