虛擬視點合成影像品質的提升

Abstract

3D虛擬視點合成系統指的是利用多個實體相機拍攝到的彩色以及深度資訊來合成任意視點的影像。然而，想要產生高品質的合成影像，必須面臨許多技術上的挑戰，包含：映射、融合、消除鬼影、填補破洞，以及減少影片中閃爍的現象。因此，本篇論文針對這些問題來提出解決的方法，包含：不可靠區域的擷取、快速逆向映射、自適應融合、利用建構出的背景填補破洞，以及強化時間軸上的一致性。其他技術像是遮蔽區域的偵測、深度圖的改善以及背景模型的建構也被納入我們的方法。最後實驗結果顯示：我們所提出的方法對於虛擬視點影像的合成，其主觀品質有相當不錯的改善。

A 3D virtual view synthesis system is to generate a virtual view at an arbitrary viewpoint from the texture and depth information of multiple reference views. There are several technical challenges in producing a high-quality synthesized view such as warping, blending, ghost artifact reduction, hole filling and flickering effect reduction. Many tools have been proposed in this study to solve these problems. They are unreliable region extraction, fast backward warping, adaptive blending, hole filling by using the constructed background models, and temporal consistency enhancement. Other techniques such as disocclusion detection, depth refinement, and background model construction are also employed in the proposed algorithm. Based on the experimental results in this study, it is found that our proposed approach performs very well on improving the subjective quality of the synthesized virtual views.