| dc.contributor | 資訊系 | |
| dc.creator (作者) | 彭彥璁 | |
| dc.creator (作者) | Peng, Yan-Tsung;Hsiao, Hung-Hsuan;Chang, Jui-Hao;Chen, Zihao;Peng, Tzu-Heng | |
| dc.date (日期) | 2026-05 | |
| dc.date.accessioned | 18-Sep-2026 15:35:15 (UTC+8) | - |
| dc.date.available | 18-Sep-2026 15:35:15 (UTC+8) | - |
| dc.date.issued (上傳時間) | 18-Sep-2026 15:35:15 (UTC+8) | - |
| dc.identifier.uri (URI) | https://ah.lib.nccu.edu.tw/item?item_id=185491 | - |
| dc.description.abstract (摘要) | Images captured through reflective surfaces often suffer from unwanted reflections that obscure the original scene. Single Image Reflection Removal aims to restore the underlying scene by removing these reflections. Unlike existing methods that rely on image decomposition or auxiliary cues, such as edge priors or linguistic cues, we propose a Multi-Exposure Reflection eLImination Network (MERLINet) for image reflection removal. MERLINet exploits hierarchical multi-scale features from images at multiple Exposure Values (EVs), derived from the input, to enhance reflection removal. A key insight is that negative EV images allow for better differentiation between transmission and reflection layers. This principle, embedded in MERLINet’s multi-scale design, simplifies the ill-posed nature of SIRR and improves reflection suppression. Extensive experiments on benchmark datasets show MERLINet performs favorably against state-of-the-art methods. | |
| dc.format.extent | 113 bytes | - |
| dc.format.mimetype | text/html | - |
| dc.relation (關聯) | 2026 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), IEEE Signal Processing Society | |
| dc.subject (關鍵詞) | Reflection Removal; Multi-Exposure Fusion | |
| dc.title (題名) | MERLINet: Multi-Exposure Reflection Elimination Network For Real-World Scenes | |
| dc.type (資料類型) | conference | |
| dc.identifier.doi (DOI) | 10.1109/ICASSP55912.2026.11464106 | |
| dc.doi.uri (DOI) | https://doi.org/10.1109/ICASSP55912.2026.11464106 | |
