Three-dimensional Super-resolution of X-ray CT Data of Rock Samples by Sparse Representation Learning

In recent years, computed tomography (CT) has been widely used during scientific drilling, providing continuous data of various rock structures such as rock layers, sedimentary layers, fractures and pores. Low-resolution CT used in drilling is insufficient to reveal the fine structures of rocks. On the other hand, X-ray CT, such as that used in the laboratory, has high resolution but is limited by the size of the sample. If the different scale-resolutions between high-resolution and low-resolution CT data can be linked, important information for multiscale analysis can be extracted. We therefore propose three-dimensional sparse super-resolution for CT data of rock samples. We show that the proposed method can reconstruct particles, veins, and texture microstructures from low-resolution three-dimensional data with super-resolution. Using multiple evaluation indices, we also demonstrate the effectiveness of the proposed method by comparing the proposed method with conventional interpolation methods.