Journal of Digital Policy (디지털정책학회지)

The Society of Digital Policy and Management (한국디지털정책학회)
권호 표지
DOI QR코드

DOI QR Code

Texture-Spatial Separation based Feature Distillation Network for Single Image Super Resolution

단일 영상 초해상도를 위한 질감-공간 분리 기반의 특징 분류 네트워크

  • Hyun Ho Han (College of General Education, University of Ulsan)
  • Received : 2023.08.08
  • Accepted : 2023.09.20
  • Published : 2023.09.28
Download PDF
⟨ Previous Next ⟩

Abstract

In this paper, I proposes a method for performing single image super resolution by separating texture-spatial domains and then classifying features based on detailed information. In CNN (Convolutional Neural Network) based super resolution, the complex procedures and generation of redundant feature information in feature estimation process for enhancing details can lead to quality degradation in super resolution. The proposed method reduced procedural complexity and minimizes generation of redundant feature information by splitting input image into two channels: texture and spatial. In texture channel, a feature refinement process with step-wise skip connections is applied for detail restoration, while in spatial channel, a method is introduced to preserve the structural features of the image. Experimental results using proposed method demonstrate improved performance in terms of PSNR and SSIM evaluations compared to existing super resolution methods, confirmed the enhancement in quality.

본 논문은 단일 영상을 이용하여 초해상도 방법을 수행하기 위해 질감-공간 영역을 분리한 뒤 세부정보를 중심으로 특징을 분류하는 방법을 제안한다. CNN(Convolutional Neural Network) 기반의 초해상도는 세부정보를 개선하기 위한 특징 추정 과정에서의 복잡한 절차와 중복된 특징 정보의 생성으로 인해 초해상도에서 가장 중요한 기준인 품질 저하가 발생할 수 있다. 제안하는 방법은 절차적 복잡성을 줄이고 중복 특징 정보의 생성을 최소화하여 초해상도 결과의 품질을 개선하기 위해 입력 영상을 질감과 공간의 두 채널로 분리하였다. 질감 채널에서는 세부정보 복원을 위해 다중스케일로 변환한 영상에 단계별 skip-connection을 적용한 잔차 블록 구조를 적용하여 특징 정제 과정을 수행함으로써 특징 추출을 개선하였고, 공간 채널에서는 평활화된 형태의 특징을 활용하여 잡음을 제거하고 구조적 특징을 유지하도록 하였다. 제안하는 방법을 이용해 실험한 결과 기존 초해상도 방법대비 PSNR 및 SSIM 성능 평가에서 향상된 결과를 보여 품질이 개선됨을 확인할 수 있었다.

Keywords

References

  1. H. Chen, X. He, L. Qing, Y. Wu, C. Ren, R. E. Sheriff & C. Zhu. (2022). Real-world single image super-resolution: A brief review. Information Fusion, 79, 124-145. DOI : 10.1016/j.inffus.2021.09.005
  2. Y. Zhang, Y. Huang, K. Wang, G. Qi & J. Zhu. (2023). Single image super-resolution reconstruction with preservation of structure and texture details. Mathematics, 11(1), 216. DOI : 10.3390/math11010216
  3. Z. Hui, J. Li, X. Gao & X. Wang. (2021). Progressive perception-oriented network for single image super-resolution. Information Sciences, 546, 769-786. DOI : 10.1016/j.ins.2020.08.114
  4. B. Niu et al. (2020). Single image super-resolution via a holistic attention network. In Computer Vision-ECCV 2020: 16th European Conference, Glasgow, UK, August 23-28, 2020, Proceedings, Part XII 16 (pp. 191-207). Springer International Publishing. DOI : 10.1007/978-3-030-58610-2_12
  5. F. Fang, J. Li & T. Zeng. (2020). Soft-edge assisted network for single image super-resolution. IEEE Transactions on Image Processing, 29, 4656-4668. DOI : 10.1109/TIP.2020.2973769
  6. M. Haris, G. Shakhnarovich & N. Ukita, (2020). Deep back-projectinetworks for single image super-resolution. IEEE Transactions on Pattern Analysis and Machine Intelligence, 43(12), 4323-4337. DOI : 10.1109/TPAMI.2020.3002836
  7. S. M. A. Bashir, Y. Wang, M. Khan & Y. Niu. (2021). A comprehensive review of deep learning-based single image super-resolution. PeerJ Computer Science, 7, e621. DOI : 10.1016/j.neucom.2019.09.035
  8. B. Liu & D. Ait-Boudaoud, (2020). Effective image super resolution via hierarchical convolutional neural network. Neurocomputing, 374, 109-116. DOI : 10.1016/j.neucom.2019.09.035
  9. C. Dong, C. C. Loy, K. He & X. Tang. (2015). Image super-resolution using deep convolutional networks. IEEE transactions on pattern analysis and machine intelligence, 38(2), 295-307. DOI : 10.1109/TPAMI.2015.2439281
  10. J. Kim, J. K. Lee & K. M. Lee. (2016). Accurate image super-resolution using very deep convolutional networks. In Proceedings of the IEEE conference on computer vision and pattern recognition (pp. 1646-1654). DOI : 10.1109/CVPR.2016.182
  11. B. Lim, S. Son, Kim, H., Nah, S., & Mu Lee, K. (2017). Enhanced deep residual networks for single image super-resolution. In Proceedings of the IEEE conference on computer vision and pattern recognition workshops (pp. 136-144). DOI : 10.48550/arXiv.1707.02921
  12. Y. Zhang, Y. Tian, Y. Kong, B. Zhong & Y. Fu. (2018). Residual dense network for image super-resolution. In Proceedings of the IEEE conference on computer vision and pattern recognition (pp. 2472-2481). DOI : 10.48550/arXiv.1802.08797
  13. Y. Huang, J. Li, X. Gao, Y. Hu & W. Lu. (2021). Interpretable detail-fidelity attention network for single image super-resolution. IEEE Transactions on Image Processing, 30, 2325-2339. DOI : 10.1109/TIP.2021.3050856
  14. Z. Lu & Y. Chen. (2022). Single image super-resolution based on a modified U-net with mixed gradient loss. signal, image and video processing, 1-9. DOI : 10.1007/s11760-021-02063-5
  15. G. Wu, J. Jiang & X. Liu. (2023). A practical contrastive learning framework for single-image super-resolution. IEEE Transactions on Neural Networks and Learning Systems. DOI: 10.1109/TNNLS.2023.3290038