Data-driven depth direction super-resolution framework for X-ray CT images of rock samples by deep learning

Rock CT images obtained from geological studies have been payed attention in earth and environmental science. Rock CT images have low resolution in depth direction while three-dimensional structure is important. Therefore, It is important to establish a method for super-resolution technologies for rock CT images in depth direction. In this study, we propose a data-driven depth direction super-resolution method for estimating X-ray CT images of three-dimensional rock samples by deep learning framework. We estimate optical flows between observable two input CT images at different depths and generate an intermediate CT image at unobservable depth by means of flow computations based on deep learning. We verify the effectiveness of the proposed method by using rock CT image data recorded in Oman drilling project. The results show that with the proposed method we can estimate intermediate CT image with better visual quality and with higher accuracy than conventional method.