摘要 在本篇論文中,我們提出一套快速的演算法來從數位照片中除去大的物件,利用將剩餘背景的資訊來將空洞填補以得到一個視覺上幾可亂真的結果。 圖像或者照片有時會包括一些噪音,污染物或者其他不想要的物件,而這些物件又佔了圖像相當大的部分。除去物體在這幅圖像內會造成一個或更多空洞。這些洞需要被填補來完成這幅圖像。傳統上,有幾種方法能解決這樣的問題,例如能用來履行這完成的複製刷子筆觸和其複合過程。不過,這些方法需要有經驗的使用者。質地綜合也能用來在填補空洞,只要其存在的地區是穩定的或是具有結構性的地方重建方法的百分之能用來透過篡改填入大規模丟失的地區。 補圖或者稱為區域填補是種方法,透過使用來自這幅圖像的剩下的區域的訊息填入這樣的一幅圖像的重要的部分。這裡我們提出利用兩種方法的一種快的填圖方法。一個Kd樹形架構是用來加速全矢量搜尋,及利用使用者的輸入來建構出一張架構地圖來減少在我們的結果和使用者的預期之間的差距。 我們的方法,基於以樣品為基底的補圖,在一般的情形裡非常有效地執行。
Abstract We presents a fast algorithm for removing large objects from digital photographs and replacing holes with remaining backgrounds which get a visually plausible result. Images or photographs sometimes may contain noises, contaminations or undesired objects which cover signi‾cant portions of the images. Removing the objects creates one or more holes in the image. These holes need to be ‾lled to complete the image. Traditionally, there are several methods such as clone brush strokes and composite processes which can be used to ful‾ll this completion. However, these methods require experienced users. Texture synthesis can also be used to complete holes where their located regions are stationary or structured tex- tures. Inpainting or region completion is a method to ‾ll in such signi‾cant portions of an image by using the information from the remaining area of the image. Here we propose a fast inpainting method that takes advantages of two approaches. A Kd-tree structure is applied to speed up full vector search and a structure map is provided by the user's input to decrease the gap between our result and the user's expectation. Our method, which is based on exemplar-based inpainting approach, performs very e±ciently in general circumstances.