@article{ART003349299},
author={Jae-Yeob Kim and Ji-Su Yoon and Hyun-Sung Jang and Seung-Wook Park},
title={Integrated Super‑Resolution Deblurring Model Using Domain Adaptation and Multi‑Task Learning},
journal={Journal of The Korea Society of Computer and Information},
issn={1598-849X},
year={2026},
volume={31},
number={6},
pages={115-124}
TY - JOUR
AU - Jae-Yeob Kim
AU - Ji-Su Yoon
AU - Hyun-Sung Jang
AU - Seung-Wook Park
TI - Integrated Super‑Resolution Deblurring Model Using Domain Adaptation and Multi‑Task Learning
JO - Journal of The Korea Society of Computer and Information
PY - 2026
VL - 31
IS - 6
PB - The Korean Society Of Computer And Information
SP - 115
EP - 124
SN - 1598-849X
AB - In this paper we propose an integrated super‑resolution–deblurring model based on multi‑task learning that employs the encoder of a deblurring network as a shared encoder. Images captured from highly dynamic platforms inevitably contain motion blur, which makes it difficult to obtain clear results when performing various image‑enhancement tasks such as super‑resolution or de‑hazing. By jointly learning super‑resolution and deblurring within a single architecture, we develop a deep‑learning model capable of producing sharp, high‑resolution images even in the presence of severe blur. To address the scarcity of training data in the infrared (IR) domain, we incorporate a domain‑adaptation technique that aligns feature distributions between the source and target domains, thereby mitigating performance degradation caused by limited IR datasets. Finally, to achieve real‑time inference (≥ 30 fps) on an on‑device platform, we compress the network using knowledge distillation. The distilled model runs at real‑time speed for 1K‑resolution inputs on an NVIDIA RTX A4500 Embedded GPU.
KW - Image Restoration;Deep Learning;Super-Resolution;Deblurring;Multitask-Learning;Domain Adaptation;Knowledge Distillation
DO -
UR -
ER -
Jae-Yeob Kim, Ji-Su Yoon, Hyun-Sung Jang and Seung-Wook Park. (2026). Integrated Super‑Resolution Deblurring Model Using Domain Adaptation and Multi‑Task Learning. Journal of The Korea Society of Computer and Information, 31(6), 115-124.
Jae-Yeob Kim, Ji-Su Yoon, Hyun-Sung Jang and Seung-Wook Park. 2026, "Integrated Super‑Resolution Deblurring Model Using Domain Adaptation and Multi‑Task Learning", Journal of The Korea Society of Computer and Information, vol.31, no.6 pp.115-124.
Jae-Yeob Kim, Ji-Su Yoon, Hyun-Sung Jang, Seung-Wook Park "Integrated Super‑Resolution Deblurring Model Using Domain Adaptation and Multi‑Task Learning" Journal of The Korea Society of Computer and Information 31.6 pp.115-124 (2026) : 115.
Jae-Yeob Kim, Ji-Su Yoon, Hyun-Sung Jang, Seung-Wook Park. Integrated Super‑Resolution Deblurring Model Using Domain Adaptation and Multi‑Task Learning. 2026; 31(6), 115-124.
Jae-Yeob Kim, Ji-Su Yoon, Hyun-Sung Jang and Seung-Wook Park. "Integrated Super‑Resolution Deblurring Model Using Domain Adaptation and Multi‑Task Learning" Journal of The Korea Society of Computer and Information 31, no.6 (2026) : 115-124.
Jae-Yeob Kim; Ji-Su Yoon; Hyun-Sung Jang; Seung-Wook Park. Integrated Super‑Resolution Deblurring Model Using Domain Adaptation and Multi‑Task Learning. Journal of The Korea Society of Computer and Information, 31(6), 115-124.
Jae-Yeob Kim; Ji-Su Yoon; Hyun-Sung Jang; Seung-Wook Park. Integrated Super‑Resolution Deblurring Model Using Domain Adaptation and Multi‑Task Learning. Journal of The Korea Society of Computer and Information. 2026; 31(6) 115-124.
Jae-Yeob Kim, Ji-Su Yoon, Hyun-Sung Jang, Seung-Wook Park. Integrated Super‑Resolution Deblurring Model Using Domain Adaptation and Multi‑Task Learning. 2026; 31(6), 115-124.
Jae-Yeob Kim, Ji-Su Yoon, Hyun-Sung Jang and Seung-Wook Park. "Integrated Super‑Resolution Deblurring Model Using Domain Adaptation and Multi‑Task Learning" Journal of The Korea Society of Computer and Information 31, no.6 (2026) : 115-124.
