視訊修補是用來移除視訊中所選擇的物件或重建影像中受損的區域,主要的挑戰是使修復後的視訊影片與原始影片有著相同的視覺品質。視訊修補必須使用強健的物件追蹤演算法,還要考慮在畫面(frame)之間的時間連續性,尤其是在攝影機移動的情況下,其考慮的因素更多。所以近年來,它已受到許多研究者的關注。本論文提出一個基於模範(Exemplar-based)方法為基礎的視訊俢補技術,首先利用哈里斯角點偵測(Harris corner detection)提取出所有畫面的特徵點,並使用仿射轉換(Affine transformation)的特性將所有的畫面疊合起來,再利用中值法建立了動態背景的全景影像(mosaic)。對於移動中的物件使用區塊匹配演算法的概念提出了一套強健的物件追蹤方法,再利用背景減法移除所有畫面的前景物件。最後使用基於模範的方法進行視訊修補。實驗證明,我們提出來的方法,不僅可以利用在動態的背景上,更可以準確的追蹤移除物件,在修補過程也能避免掉影像中線性結構的遺失,不會產生模糊,因此,修補後的視訊可以達到一定的品質。
Video inpainting is a technique to remove selected objects or reconstruct damaged area of a video automatically. The main challenge is the restored sequence should have the same visual quality as the original. It not only needs explore a robust object tracking algorithm, but also need take into account the temporal continuity among video frames, especially in the case that the video has camera motions. Hence, video inpainting has drawn much attention in the past few years. This paper proposed a video inpainting scheme based on exemplar-based method. Firstly, the Harris corner detection is used to extract the feature of the frames and Affine transformation is applied to stitch all frames up, and then the median method is used to establish a dynamic background mosaic image. After that, a new robust object tracking method is proposed for moving object using the concept of optical flow. Next, background subtraction is adopted to remove the foreground moving object among all fames. Finally, the video is repaired by the exemplar-based approach. Experimental results demonstrate the proposed method not only can be using in the dynamic background case, but also accurately tracked at remove selected objects. The inpainting process can avoid producing blur when the linear structure in image is missed. Thus, the repaired video can achieve a certain degree quality.