- 學位論文
自由視角電視與3D視訊的視角生成與空洞補償
View Generation & Hole-Filling in FTV and 3D Video
摘要
在本篇論文中所完成的是討論一個新興的自由影像合成 (Free View Synthesis) 方法，應用在自由視點的電視(FTV)的深度影像繪圖法(DIBR)。其中一個關於影像合成概念的問題為影像資訊遮蔽在原始影像的物體，可能會在合成的影像中變得可見，特別是在推測視角(extrapolated views)，而用來做洞補償未知紋理缺少第二個影像的時候。二個新的方法用來補償在合成影像中的未知紋理在本篇論文中提出。在第一個方法中我們提出了利用軌道(trajectory)的方式，已建立了經典的渲染過程。這包含了在DIBR中詳細的座標投影法。在我們的研究我們展示了在原始影像中每一列的像素當使用DIBR映射到合成影像上的行為。而在合成影像上像素的則使用迴歸分析來預測未知區域的圖像軌跡。另外一個方法提出一種最佳化系統使得在合成未知的紋理，會去考慮到空間域與時間域的相關性。實驗結果展示了合成的影像中重要的物件有最小的幾何失真，而在人眼主觀上的評比也優於目前最先進的方法。
並列摘要
The work accomplished in this thesis discusses a novel free view synthesis (FVS) method for free view point television (FTV) using Depth Image Based Rendering (DIBR). One of the problems with the view synthesis concept is the fact that image information that is occluded in the original view may become visible in the synthesized view, especially in extrapolated views where the advantage of second view for filling unknown texture is absent. Two novel approaches for filling the unknown texture in the synthesized view are proposed. In one approach we propose to make use of the trajectory, established by the classical rendering procedure. It includes detailed study of coordinate projection method in DIBR. In our study we show that how the pixels of each row in the original view behave when mapped on the synthesized view using the DIBR. The behaviour of the pixels in the synthesized view is regressed to predict the trajectory over the unknown region. The other approach proposes an optimized framework that synthesizes the unknown textures by taking into account the spatial and temporal correlation. Experimental results show that the synthesized views have minimized geometric distortions with significant objective and subjective gains in comparison to the state-of-art methods. The thesis concludes by describing features that make the proposed approach very suitable for extrapolation of virtual view synthesis. Meanwhile, the proposed framework has the flexibility of shifting to the interpolation scenario.