This paper presents two algorithms to distinguish real faces from flat photos in stereo camera images. The first algorithm based on geometric relations between certain face landmarks. All points of the face photo will lie on the same plane, but for the real face - no. The second uses the generated face relief pattern. The face is 2D or 3D dependent on the similarity of the generated normalization pattern to the pretrained template. Proposed approaches allow obtaining a number of advantages of 3D face recognition systems with minimal costs to modernization the existing 2D systems. Some recommendations and limitations to the usage of the considered approaches are determined.