We are interested in the applications of invariant theory to computer vision problems. A survey and clarification of the different invariant calculation methods are detailed in our extended technical report. In this paper, we concentrate on 3-D invariants from pairs of images instead of invariants related to the planar projective transformations from monocular image that have been already largely studied by many people. Especially invariants of different configurations coming from different combinations of 3-D points, 3-D lines and conics are under consideration, and the stabilities of the invariants computed differently are compared. We conclude by showing to what extent these invariant measures can be used for recognition and indexing.