臺灣大學: 資訊網路與多媒體研究所吳家麟曾翊寧Tseng, Yi-NingYi-NingTseng2013-03-222018-07-052013-03-222018-07-052011http://ntur.lib.ntu.edu.tw//handle/246246/251160淺景深是一種利用淡化背景雜訊來強化主題的拍攝手法,較常使用在人像或是近物攝影,此種特效照片的視覺效果優異,因此較受人們所喜愛。不過一般低價位消費型數位相機由於硬體構造的限制,無法產生類似高價位單眼相機拍攝的淺景深的照片。近年來已有內建式相機軟體(如Sony Cyber-shot WX5 and TX9)或是後製影像處理工具(Photoshop)利用軟體模擬照片中的光學性質。但由於以上方式無法藉由單張照片得到深度資訊,使得模擬淺景深的效果失真度較高,基於以上動機,我們開發出一套使用者互動式介面可方便而準確的取得深度資訊,再利用即時重新聚焦演算法並在模擬淺景深時加入散景的效果,進而建構出一套不僅快速且效果逼真的數位照片淺景深模擬系統。Shallow focus is a kind of photographic technique, which means taking a picture with a small depth of field (DOF) so that the objects outside of the DOF will be blurred to highlight the subject inside of the DOF. Common digital cameras, unlike the Digital Single Lens Reflex Camera (DSLR), cannot produce this special vision effect due to its limitations in aperture and focal length. Commercial softwares and even some common digital cameras are endued with built-in shallow focus simulation programs. However, the rendering quality of the existing tools is not very well due to several problems: 1) boundaries of the focused objects are often over smoothed caused by poor auto-generated object segmentation result, 2) the blurring effect usually looks fake since the element of image formation principle, i.e. the depth of the object, is not properly utilized in the rendering process, and 3) even though some refocusing methods model the blurring effect with depth information, it is hard to extract accurate depth map from a single image. Therefore, we develop a friendly user interface which can interactively achieve reliable depth labeling based on the techniques of semi-automatic object segmentation and 3D modeling. Moreover, to realize shallow focus effect efficiently, we apply a real-time gather-based DOF rendering method and modify it to be capable of simulating the bokeh effect.2875614 bytesapplication/pdfen-US散景景深即時互動式使用者介面深度資訊Depth of field, bokeh effectshallow focusreal-timeImage defocusing一套利用深度標註即時生成具散景景深效果的互動式軟體Realtime Depth of Field Rendering with Bokeh Effect via Interactive Depth Labelingthesishttp://ntur.lib.ntu.edu.tw/bitstream/246246/251160/1/ntu-100-R98944028-1.pdf