Sequential Degradation-Consistent Parameter Scheduling for Image Restoration
- DOI
- 10.2991/978-94-6239-828-3_4How to use a DOI?
- Keywords
- Image restoration; Reinforcement learning; Parameter scheduling; Degradation consistency; Deep Q-network
- Abstract
Abstract. Image restoration pipelines often depend on degradation-dependent parameters, such as denoising strength and data-consistency weights, that are fixed before inference or selected through computationally expensive grid search. This paper proposes a lightweight sequential parameter scheduler for degradation- consistent image restoration. The restoration process is formulated as a finite- horizon Markov decision process, where a compact Deep Q-Network selects denoising and data-consistency parameters at each restoration step based on non-reference state features extracted from the degraded observation and inter- mediate restoration output. The reward balances reconstruction improvement, structural similarity, runtime cost, and degradation-consistency residuals, while clean-reference metrics are used only during training and not during test inference. Experiments on Set12-style AWGN simulations under πβββ{15, 25, 50} show that the proposed scheduler consistently improves degraded inputs and random sequential scheduling while substantially reducing oracle-style grid-search evaluations. However, the current policy does not yet consistently outperform stronger practical baselines such as fixed parameter scheduling and one-shot prediction. These results indicate that DQN-based sequential scheduling is a feasible and efficient mechanism for adaptive restoration, but further full-scale benchmark validation is still required.
- Copyright
- Β© 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.
Cite this article
TY - CONF AU - Syahrul Budi Rahmadan AU - Muhamad Rizqi Duha Pramudya AU - Ikhlasul Amal AU - Wayan Firdaus Mahmudy AU - Novanto Yudistira PY - 2026 DA - 2026/11/04 TI - Sequential Degradation-Consistent Parameter Scheduling for Image Restoration BT - Proceedings of the 2026 11th International Conference on Sustainable Information Engineering and Technology (SIET 2026) PB - Atlantis Press SP - 32 EP - 42 SN - 2589-4900 UR - https://doi.org/10.2991/978-94-6239-828-3_4 DO - 10.2991/978-94-6239-828-3_4 ID - Rahmadan2026 ER -