The lamina cribrosa (LC), a 3D porous structure of the eye in which optic fibers pass to reach the brain, is critically involved in the diagnosis and pathogenesis of glaucoma, a leading cause of blindness. However, segmenting the LC pores in 3D images is a task that has rarely been investigated, mainly because of the very low signal to noise ratio. To address this problem, we propose a fully automatic method to register and fuse two 3D SD-OCT volumes acquired in orthogonal directions, in order to improve the OCT image quality and thus facilitate the pore detection. Our method relies on a priori knowledge about en-face images to enhance useful pore information and initialize the process. Moreover, the optimization of the cross-correlation function is carried out in several stages in order to ensure the robustness of the process and the accuracy of the result. The quantitative evaluation shows that the proposed method is efficient to register the two 3D orthogonal OCT volumes and the annotated pores, with a distance between aligned pores around 3 pixels (below the pore size).