One of the main research subject in Computer Vision and Photogrammetry is the correspondence problem or image matching. The interest in this subject can be justified once it is considered as an ill-posed problem, and a robust solution is difficulty to be found.In the case where the images are not in a normal disposition, i.e., affected by convergence, rotation and scale difference, the difficulty is even increased. So, in this work an approach based on relaxation labeling is proposed, where the relative orientation and points correspondences are automatically solved for non-normal pairs of images. In the proposed algorithm, multiples metrics are considered on both, similarity and compatibility computation. Among the metrics used, one of them is the angular relation between the neighborhood at primitive space, and epipolar constraints, via the volume of Matching Parallelepiped -MP. The volume of the MP can be computed when the relative orientation is available, and it was shown that the MP volume is related to the epipolar geometry. Therefore, it is not necessary to compute the epipolar line equations and distances between the candidates to the epipolar lines. Experiments with synthetic and real images indicate that even for situations where scale differences, rotation and convergence are presented, the relative orientation parameters are recovered and most of the correspondences are correctly found.