隨著動畫技術以及市場的成熟及進步,全世界每年出產許多的動畫影片。根據日本動畫協會的統計,從2004年起,日本每年都產出至少200部的動畫影片。對於動漫迷來說,每天或每星期他們迫不及待的希望趕快看到每部新出品的影片內容。因此,如何有效瀏覽大量的動畫成為一個重要的問題,在本論文我們提出一套可以將動畫轉換為漫畫的系統,除了可以加快瀏覽的速度之外,同時增加了瀏覽的樂趣。我們首先偵測鏡頭切換,並為每個鏡頭選擇一些畫面來表示,接著進行場景偵測。針對個別的場景,我們設計一個最佳化架構來安排那些畫面要放到同一漫畫的頁面中。配置的方法有許多可能性,且每個畫面都是必須連續的,所以在本論文我們採用基因演算法,經由演化的過程得到一個最好的配置。得到了安排的頁面中有哪些畫面後,接著是安排這些畫面要怎樣排版。首先我們設計出許多版面,並且經由計算畫面與版面的相似性,選擇最好的版面。選完版面後,我們在畫面中選取適當的區域,然後將畫面縮放貼到對應的畫格。最後我們就可以得到由原始動畫影片產生的漫畫。 在實驗中我們計算場景偵測的精準度、基因演算法的遞迴狀況,以及進行使用者的主觀實驗。在此我們以轉換後的漫畫,原始的動畫影片以及由商業軟體KMPlayer所產生的預覽圖做比較,根據使用者的回饋,大部分的使用者肯定我們的效果並且展現高度的興趣。
With the advances of animation technology and industries, many animations are produced every year. According to the latest statistics released by the Association of Japanese Animations, more than 200 animations produced every year from 2004 only in Japan. For animation fans, they keep their eyes on the newly-released episodes every day or every week. How to efficiently browse large volume of animation, therefore, becomes an important problem. In this thesis, we propose a media transforming system that transforms animations into comics. This system can not only increase browsing speed but also increase browsing experience. We first conduct shot change detection, select keyframes from each shot, and conduct scene detection. We devise a method to arrange frames into pages, and then select the most appropriate layout for presentation. We select appropriate region of each keyframe to stick on the targeted cell of the layout. Finally, we can generate comics that appropriately display information of the original animation. In experiments, we show performance of scene boundary detection and the iteration process of the optimization framework. In addition, we design a user study to evaluate how users feel about the generated comic pages. The user study show very promising results.