利用兩張照片影像對齊與融合之物體移除技術

Abstract

攝影時，常常會遇到物體闖入照片的情形，這導致攝影者原先的構圖被破壞，且照片的美感降低。然而，因某些特定因素，攝影者依然希望保留此張被破壞構圖的照片，且希望將闖入的物體去除。因此，我們提出了一個方法，以解決此種問題。為了去除使用者所選定的不要物體，前人之方法藉由抽取被破壞照片本身的資訊或搜索網路上相似之照片，以填補不要的物體。反之，我們專注在使用智慧型手機之使用情境上且對兩張由相同場景收集來的照片做處理。我們提出的方法可分為兩個步驟：影像對齊與影像融合。藉由估計兩張照片之間的座標轉換，影像對齊的技術可校正兩張照片中不同的視野，以便在來源照片中取得對應的區域，用以填補欲移除之區域。得到對應區域的影像後，影像融合技術消弭兩張照片因不同光照環境影響之色差。經由此兩步驟的處理，我們所提出的方法，可以消除照片中意外闖入的物體，並填入適當的影像。

Often when photographers take pictures, some undesired objects rush into pictures. It will break the framing of these pictures. However, photographers still want to keep these pictures for some reasons. Hence, we propose a method to remove the undesired object in pictures. To remove the undesired object specified by users, some previous works draw the information from the image itself and other previous works search the similar scene from internet to complete the undesired object. Instead, we focus on the scenario of using smartphone and process with two pictures collected from the same scene. Our method consists of two stages: image alignment and image blending. By estimating the homography between images, image alignment calibrates the difference of geometry between images. Therefore, we can locate the source patch, used to complete the undesired region. After image alignment, image blending smoothly spread the discrepancies of photometry between images under different lighting conditions. After these two steps, our work removes the rush-in objects and fills in the appropriate images.