Image Super Resolution Enhancement Using A Hybrid Model Framework

DOI

10.2991/978-94-6239-674-6_39How to use a DOI?

Keywords

Super Resolution; Convolutional Neural Network (CNN); Transformer; Generative Adversarial Network (GAN); Fusion Network; Hybrid Framework

Abstract

The goal of Image Super-Resolution is to obtain high-resolution images from low-resolution images, but it is still difficult to obtain both structure accuracy and perceptual sharpness. CNN models have been shown to restore edges and textures, but results can often be overly smooth. Transformer architectures are able to model long-range spatial dependency effectively, but can lead to structural inconsistencies. Likewise, GAN-based image enhancement gives a high degree of perceptual realism, but can often produce undesirable artifacts when used alone. In response to these limitations, this paper proposes a new approach to SR by presenting a hybrid Super-Resolution framework based on a training pipeline that includes Convolutional Neural Networks, Transformer encoders, and a Generative Adversarial Network. The process of training the model progresses through four stages: baseline CNN reconstruction, global context refinements from the Transformer-oriented architectural attention, adversarial learning for perceptual textures, and finally, a Fusion Network that adaptively blends outputs from the three models. The DIV2K dataset is employed for training and evaluation. The hybrid framework composites complementary variations of the individual architectures in an effort to achieve improvements in detail preservation, structural coherence, and perceptual realism compared to their respective baselines employing CNN-only, Transformer-only, or GAN-only approaches to super-resolution. Overall, we believe the proposed framework has great potential for real-world uses, such as medical imaging, satellite observation, and visual restoration applications, when accuracy and reliability are paramount.

Copyright

© 2026 The Author(s)

Open Access

Open Access This chapter is licensed under the terms of the Creative Commons Attribution-NonCommercial 4.0 International License (http://creativecommons.org/licenses/by-nc/4.0/), which permits any noncommercial use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license and indicate if changes were made.

Download article (PDF)

TY  - CONF
AU  - Uday Sharma
AU  - Kanika Mahajan
AU  - Sanjana Kharbanda
AU  - Shubham Kathuria
AU  - Swapnil Kaushal
PY  - 2026
DA  - 2026/05/28
TI  - Image Super Resolution Enhancement Using A Hybrid Model Framework
BT  - Proceedings of the International Conference on Sustainable Computing and Artificial Intelligence (ICSCAI 2025)
PB  - Atlantis Press
SP  - 472
EP  - 485
SN  - 2352-5401
UR  - https://doi.org/10.2991/978-94-6239-674-6_39
DO  - 10.2991/978-94-6239-674-6_39
ID  - Sharma2026
ER  -