摘要
摳像(Image Matting)是一個從輸入的影像或是影片中，擷取出前景的過程。摳像在電影工業以及影像處理中都是一項重要的技術，可以用於製作電影特效以及影像合成。所以，如何分離出完整的前景，就成為了一個重要的研究項目。在處理複雜背景的影像時，大部分的方法都會需要使用者提供額外的資訊，來標示何處是前景，何處是背景，有些地方則是混雜了前景與背景的區域，這時，使用者標示的資料，通稱為三原圖(Trimap)，就顯的相當重要。 本研究旨在將整個摳像的過程，包含產生三原圖，計算前景區域的部份，全部使用深度學習方法，成為全自動的過程，使用者只需輸入一張圖像，程式即可自動判斷前景的區域，並將前景計算出來。本研究也提出了一個新的深度學習摳像方法。研究結果顯示，在背景不單一的情況下，本研究仍能夠自動產生三原圖，且本研究提出的摳像方法也勝過其他的摳像方法。
並列摘要
Alpha matting is a method to extract the foreground from an input image or a video. It is an important technology in the film industry and image processing, which can be used to make movie effects and image synthesis. Therefore, how to separate the complete foreground has become an important research project. When input image has complex background, most of previous methods require users to provide additional information to indicate which part of the image is foreground region, which is the background region, and the other is uncertain region. The additional information, commonly known as the trimap, is quite important to the input image has complex background. The purpose of this study is to make the whole alpha matting process, including generating trimap and calculating the alpha matte, become a fully automatic process by deep learning method. The user can only input the image and our system can automatically calculate. the area of the foreground. This study also proposes a new DL-based image matting method. The research results show that when there are multiple background regions, the research can still automatically generate the trimap, and the image matting method proposed in this study is better than other methods.