DC Field | Value | Language |
---|---|---|
dc.contributor.advisor | Kim, Min Hyuk | - |
dc.contributor.advisor | 김민혁 | - |
dc.contributor.author | Kim, Jungwoo | - |
dc.date.accessioned | 2023-06-26T19:31:21Z | - |
dc.date.available | 2023-06-26T19:31:21Z | - |
dc.date.issued | 2023 | - |
dc.identifier.uri | http://library.kaist.ac.kr/search/detail/view.do?bibCtrlNo=1032964&flag=dissertation | en_US |
dc.identifier.uri | http://hdl.handle.net/10203/309513 | - |
dc.description | 학위논문(석사) - 한국과학기술원 : 전산학부, 2023.2,[iv, 24 p. :] | - |
dc.description.abstract | Spatial resolution of an image sensor has increased significantly, allowing us to capture additional information, such as spatially-varying multiple exposures. Quad-Bayer patterned sensors have become popular, enabling snapshot HDR imaging. However, the quad-Bayer pattern compromises spatial resolution for multiple exposures and colors in results. Also, the general quad-Bayer sensor controls exposures by gain rather than the actual exposure time, and thus the extension of dynamic range has been insignificant. While time-varying multiple exposures are desirable for high-quality HDR imaging, it is inapplicable to dynamic scenes in general, requiring an additional deghosting method. In this work, we propose a snapshot HDR demosaicing method, which takes time-varying multiple exposures as input and jointly solves both demosaicing and deghosting problems from the quad-Bayer patterned input. Our method consists of a feature-extraction module to handle mosaiced multiple exposures and a U-net style multiscale transformer module to register spatial displacements of multiple exposures and colors to obtain a high-quality HDR image from a quad-Bayer RAW image. We create a dataset of quad-Bayer sensor input with varying exposures and colors and train our network using our dataset. Results validate that our method outperforms baseline HDR reconstruction methods with both synthetic and real datasets. | - |
dc.language | eng | - |
dc.publisher | 한국과학기술원 | - |
dc.subject | HDR imaging▼aDemosaicing▼aImage restoration▼aTransformer | - |
dc.subject | HDR 이미징▼a디모자이킹▼a이미지 복원▼a트랜스포머 | - |
dc.title | Demosaicing a time-varying exposures array for snapshot HDR imaging | - |
dc.title.alternative | 스냅샷 HDR 이미징을 위한 다중 노출시간 센서 디모자이킹 기술 | - |
dc.type | Thesis(Master) | - |
dc.identifier.CNRN | 325007 | - |
dc.description.department | 한국과학기술원 :전산학부, | - |
dc.contributor.alternativeauthor | 김정우 | - |
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.