請用此 Handle URI 來引用此文件:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/70422
標題: | 以無須動作校準之梯度域亮度融合進行高動態範圍去鬼影 HDR Deghosting Using Motion-Registration-Free Fusion in Luminance Gradient Domain |
作者: | Cheng-Yeh Liou 劉承曄 |
指導教授: | 盧奕璋(Yi-Chang Lu) |
關鍵字: | 高動態範圍去鬼影,高動態範圍影像,多曝光融合, HDR deghosting,high dynamic range imaging,multi-exposure fusion, |
出版年 : | 2018 |
學位: | 碩士 |
摘要: | 利用多張不同曝光程度之影像合成高動態範圍(High Dynamic
Range, HDR) 影像已被廣泛使用在許多相機上,但是當被拍攝的物 體有移動時,HDR 影像合成會變得相當困難。HDR 去鬼影(HDR Deghosting) 即是為了解決這類問題而產生的方法。但一般的HDR 去 鬼影方法常因為運算時間太長或結果不好等因素,造成現在還未廣泛 在一般使用者間普及。本論文提出一嶄新的HDR 去鬼影方法,主要特 點有二。第一,對一般HDR 去鬼影方法而言,動作校準是一步重要且 耗時的步驟,本論文提出無須動作校準的方法,省略此步驟後可以大 幅加速整體運算。第二,本論文提出新的移動區域偵測演算法,利用 曝光轉換後進行顏色轉換的方法,可使影像間之移動區域,有更高的 曝光程度一致性,彷彿這些影像就是在同一個曝光程度中被拍攝,因 此能在移動區域偵測中得到更好的偵測結果。總結而言,本論文提出 新穎與高速之HDR 去鬼影方法,對於大部分包含移動物體之輸入影 像皆能產生高品質的HDR 結果。另外,高速運算之特性使得本方法非 常適合運用於一些未來應用,如HDR 影片、即時HDR 預覽等等。 Synthesizing a high dynamic range (HDR) image from multiple images with different exposure times has been widely adopted in the field of digital photography; however, it could be challenging when there are moving objects in the scenes. HDR deghosting methods are those which not only create HDR images but also remove ghosting effects caused by moving objects. Unlike static HDR imaging methods, HDR deghosting methods are not often used by professional or amateur photographers because the processing time is long and final HDR qualities are far from satisfactory. In this thesis, without applying time-consuming motion registration required by other HDR methods, the proposed algorithm has shorter running time. Additionally, a new moving region detection method is introduced. By performing color transfer right after exposure transfer, the input images for moving object detection stage show better exposure consistency as if they are captured under the same camera setting, thus producing better moving region detection result. In conclusion, this thesis demonstrates a novel and time-efficient algorithm which produces high quality HDR images for the scenes that have moving objects. Besides, the time-efficient feature makes it suitable for further HDR applications like HDR video or real-time preview of HDR images. |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/70422 |
DOI: | 10.6342/NTU201802781 |
全文授權: | 有償授權 |
顯示於系所單位: | 電子工程學研究所 |
文件中的檔案:
檔案 | 大小 | 格式 | |
---|---|---|---|
ntu-107-1.pdf 目前未授權公開取用 | 21.88 MB | Adobe PDF |
系統中的文件,除了特別指名其著作權條款之外,均受到著作權保護,並且保留所有的權利。