COMPARATIVE ANALYSIS OF A COMPACT RESIDUAL CONVOLUTIONAL MODEL AND LINEAR REGRESSION FOR IMAGE SUPER- RESOLUTION
DOI:
https://doi.org/10.5281/zenodo.23320605Abstract
The article evaluates a compact residual convolutional model for fourfold image super-resolution under conditions of limited training data. A total of 19 JPEG images were divided into 10 training, 3 validation, and 6 test images. The model, consisting of 433 parameters, was compared with bicubic interpolation and Ridge regression. Extending the training period from 40 to 120 epochs increased the mean test PSNR across three random initializations from 29.7675 to 29.7744 dB. The extended model achieved an SSIM value of 0.83569, while the Ridge model obtained a PSNR of 29.8222 dB and an SSIM of 0.83424. In the controlled variants, the window size, hidden-layer width, and input centering were investigated. Paired bootstrap confidence intervals and CPU processing time were also calculated. The results demonstrate a criterion-dependent trade-off between the evaluated approaches.Keywords
image super-resolution, residual learning, convolutional model, Ridge regression, ablation study, PSNR, SSIMReferences
Dong, C., Loy C., C., He, K., & Tang, X. (2015). Image Super-Resolution Using Deep Convolutional Networks. 2015, qayta ko‘rilgan preprint. arXiv:1501.00092v3. https://arxiv.org/abs/1501.00092.
Dong, C., Loy C., C., & Tang, X. (2016). Accelerating the Super-Resolution Convolutional Neural Network. 2016. https://arxiv.org/ abs/1608.00367.
Lim, B., Son, S., Kim, H., Nah, S., & Lee K., M. (2017). Enhanced Deep Residual Networks for Single Image Super-Resolution. 2017. https://arxiv.org/abs/1707.02921. //arxiv.org/abs/1707.
Wang, X., Xie, L., Dong, C., & Shan, Y. (2021). Real-ESRGAN: Training Real-World Blind Super-Resolution with Pure Synthetic Data. 2021. https://arxiv.org/abs/2107.10833.
Blau, Y., & Michaeli, T. (2018). The Perception-Distortion Tradeoff. CVPR, 2018. https://arxiv.org/abs/1711.06077. //arxiv.org/abs/1711.
Wang, Z., Bovik A., C., Sheikh H., R., & Simoncelli E., P. (2004). Image Quality Assessment: From Error Visibility to Structural Similarity. IEEE TIP, 13(4), 600-612, 2004. https://ece.uwaterloo.ca/~z70wang/publications/ssim.html. //ece.uwaterloo.ca/~z70wang/publications/ssim.
Esmaeilzehi, A. (2022). Development of Deep Learning Techniques for Image Super Resolution. PhD dissertatsiyasi. Concordia University, 2022. https://spectrum.library.concordia.ca/id/eprint/990271/.
Zhu, J. (2023). Deep neural networks for medical image super-resolution. PhD dissertatsiyasi. University of Cambridge, 2023. DOI: 10.17863/CAM.101535. https://doi.org/10.17863/CAM.101535. https://doi.org/10.17863/CAM.101535
Mamatov N., S., Erejepov K., K., Zarat, S., & Najmiddinov A., S. (2024). Tasvirlardagi obyektlar sifatini oshirish algoritmlari. Новый Узбекистан: наука, образование и инновации, 2024, 179-182-b. https://inlibrary.uz/index.php/new-uzbekistan/ article/view/32153. //inlibrary.uz/index.
Mamatov N., S., Jalelova, M., Erejepov, K., Samijonov, A., & Najmiddinov, A. (2024). Rentgenografik tasvirlarga dastlabki ishlov berish dasturiy majmuasi. Новый Узбекистан: наука, образование и инновации, 2024, 172-175-b. https://inlibrary. uz/index.php/new-uzbekistan/article/view/32151. //inlibrary. uz/index.
Downloads
Published
How to Cite
Issue
Section
