近年來,立體3D影像編輯在電腦圖學領域裡被廣泛地受到重視,編輯立體3D影像與一般的影像編輯不同,除了針對影像中的內容去做編輯之外,也需要考慮到影像中內容的深度。在這篇論文中,我們提出了一個結構性立體3D影像的編輯架構,這個架構額外參考了立體3D影像中所隱含的深度資訊,能讓使用者針對影像內容做編輯調整,並且能產生符合使用者編輯目標的結果。這個架構可以達到立體3D影像的物件仿製和立體3D影像的內容生成。對於物件的仿製,本論文提出的方法會同時處理物件形狀修正以及顏色融合,讓結果不論在深度或是顏色上都能夠自然。對於影像內容生成,我們提出了一個多元且可靠的影像生成技術,可以達成不同的使用者編輯目標,例如:影像縮放(image retargeting)、影像修補(image inpainting)、紋理生成(texture synthesis)。
Recently, stereoscopic 3D image editing has attracted a great deal of research interest in the computer graphics community. Different from conventional 2D image editing, it usually requires adjustment of the image content as well as the perceived depth. In this dissertation, we present a framework for structural stereoscopic 3D image editing. The framework utilizes the additional information in stereoscopic 3D image, depth structure, to manipulate the image content with respect to the user-specified editing goals. We first adapt the framework to achieve two editing operations: stereoscopic 3D image cloning and stereoscopic 3D image synthesis. The proposed stereoscopic image cloning technique performs both shape adjustment and color blending so that the stereoscopic composite is seamless in both the perceived depth and color appearance. Additionally, we introduce a versatile and robust stereoscopic image synthesis technique for synthesizing stereoscopic images with respect to editing constraints used in different editing applications (eg, image retargeting, image inpainting, texture synthesis). We demonstrate several challenging cases to show the success of our proposed techniques.