Orbital bone fractures usually occur in the thin bone of the orbital medial wall and the orbital floor. To reconstruct the fractured orbital bone, it is necessary to create a 3D model of the orbital bone by accurately segmenting the orbital bone from the facial CT image. However, since the slice thickness of the facial CT is two to three times larger than the pixel spacing, the segmentation accuracy of the thin bone is degraded and the 3D model shows an aliasing effect. In this paper, we propose an orbital bone segmentation method by improving the inter-slice resolution of the facial CT. First, to improve the inter-slice resolution, intermediate slices are generated by considering the orbital bone structure information in two orthogonal planes using a 2.5D CNN-based network. Second, to consider the wide intensity range of orbital bone, the orbital bone is divided into cortical bone and thin bone area and segmented using a 3D ensemble network. Third, to validate the effect of the inter-slice resolution improvement, the orbital bone is segmented using the image with improved inter-slice resolution. Experimental showed that the PSNR, SSIM, VIF of the proposed method was 30.80, 0.9486, 0.6238, respectively in the whole orbital bone area. In the orbital bone segmentation results of the generated images using the OB-SRNET, the Dice similarity coefficient was 96.50% in the whole orbital bone. In the thin bone of the medial wall and the orbital floor, the recall of the proposed method was 65.25% and 57.98% respectively, which is improved by 15.17% and 43.95% respectively compared to the images with the slice thickness of 2mm. The proposed method generated intermediate slice images similar to the original slice images and improved the orbital bone segmentation performance of the images with the large slice thickness. The proposed method can be used to provide customized bone plates for the reconstruction of fractured orbital thin bones in craniomaxillofacial surgery.