Super Resolution Convolutional Neural Network Models for Enhancing Resolution of Rock Micro-CT Images (1904.07470v1)

Abstract: Single Image Super Resolution (SISR) techniques based on Super Resolution Convolutional Neural Networks (SRCNN) are applied to micro-computed tomography ({\mu}CT) images of sandstone and carbonate rocks. Digital rock imaging is limited by the capability of the scanning device resulting in trade-offs between resolution and field of view, and super resolution methods tested in this study aim to compensate for these limits. SRCNN models SR-Resnet, Enhanced Deep SR (EDSR), and Wide-Activation Deep SR (WDSR) are used on the Digital Rock Super Resolution 1 (DRSRD1) Dataset of 4x downsampled images, comprising of 2000 high resolution (800x800) raw micro-CT images of Bentheimer sandstone and Estaillades carbonate. The trained models are applied to the validation and test data within the dataset and show a 3-5 dB rise in image quality compared to bicubic interpolation, with all tested models performing within a 0.1 dB range. Difference maps indicate that edge sharpness is completely recovered in images within the scope of the trained model, with only high frequency noise related detail loss. We find that aside from generation of high-resolution images, a beneficial side effect of super resolution methods applied to synthetically downgraded images is the removal of image noise while recovering edgewise sharpness which is beneficial for the segmentation process. The model is also tested against real low-resolution images of Bentheimer rock with image augmentation to account for natural noise and blur. The SRCNN method is shown to act as a preconditioner for image segmentation under these circumstances which naturally leads to further future development and training of models that segment an image directly. Image restoration by SRCNN on the rock images is of significantly higher quality than traditional methods and suggests SRCNN methods are a viable processing step in a digital rock workflow.

• The paper demonstrates how SRCNN models boost image resolution, achieving a 3-5 dB PSNR gain compared to basic bicubic interpolation.
• The study employs SR-Resnet, EDSR, and WDSR on a 2000-image DRSRD1 dataset to optimize the balance between field of view and resolution.
• The enhanced image quality improves pre-processing for segmentation tasks, paving the way for integrated super-resolution and segmentation approaches.

Super Resolution CNN Models for Enhancing Resolution of Rock Micro-CT Images

The paper "Super Resolution Convolutional Neural Network Models for Enhancing Resolution of Rock Micro-CT Images" by Wang YD, Armstrong R.T., and Mostaghimi P. investigates the application of Single Image Super Resolution (SISR) techniques based on Super Resolution Convolutional Neural Networks (SRCNNs) to enhance micro-computed tomography (micro-CT) images of porous rocks, specifically sandstones and carbonates. Employing models such as SR-Resnet, Enhanced Deep SR (EDSR), and Wide-Activation Deep SR (WDSR), the research showcases promising improvements in image quality.

Technical Assessment and Numerical Findings

The paper capitalizes on the Digital Rock Super Resolution 1 (DRSRD1) dataset, which comprises 2000 high-resolution (HR) images synthetically downsampled by a factor of 4. This downscaling, a recurring challenge in digital rock physics, aims to balance the field of view and resolution limits of current scanning methodologies. The authors observed that SRCNN models, when compared with basic bicubic interpolation, yield a significant enhancement, reflected by a 3-5 dB increase in Peak Signal-to-Noise Ratio (PSNR). The WDSR model, in particular, demonstrates superior performance with efficient parameter usage, delivering slightly better results compared to its counterparts while maintaining computational efficiency.

The research evaluates SRCNN's potential not only in producing higher resolution images but also in preconditioning these images for subsequent segmentation tasks, a pivotal stage in digital rock analysis. By minimizing intra-granular noise and maintaining edge accuracy, the SRCNN models establish a viable pre-processing step for improving segmentation outcomes.

Implications and Future Directions

In terms of practical implications, the discussed methodologies advance the digital rock workflow by alleviating the resolution-coverage tradeoff inherent in micro-CT imaging. The success of SRCNNs in mitigating high-frequency noise suggests a potential pivot towards models that integrate both super-resolution and segmentation functions seamlessly. Moreover, the paper highlights the importance of dataset augmentation to simulate realistic imaging conditions, thereby improving model generalizability.

Theoretically, the findings set a precedent for applying deep learning paradigms to geophysical imaging challenges. The robustness of SRCNNs against noise emphasizes their adaptability beyond mere resolution enhancement, presenting an intriguing avenue for future research in 3D image reconstruction within the domain of digital rock physics.

In alignment with the scope of digital imaging, the paper underscores the merit of further exploration into 3D-specific super-resolution networks, which would capture depth information often disregarded in current 2D-based models. Additionally, coupling super-resolution techniques with direct segmentation models could revolutionize image processing pipelines, especially in non-photorealistic contexts where precision and feature preservation are paramount.

In conclusion, the paper presents a comprehensive paper into SRCNN applications for micro-CT images of rocks, highlighting both performance metrics and methodological advancements. The work provides a strong foundation for future innovations aimed at refining digital rock analysis techniques, with implications extending across various subsurface imaging applications.