隨著時代演進，科技日益進步，人們對於電子產品以及３Ｃ產品越來越依賴。從以前人們拿著傻瓜相機或是具有底片的相機拍照的時代，轉變成現在的人手一台數位相機、數位單眼相機或是智慧型手機，由於3C產品的便利性，一般人即可拍出專業照片。但一張尺寸固定的照片，要放在不同比例的裝置會造成圖片內容扭曲。就好比iPhone為例，iPhone的螢幕尺寸為3比2，如果要將比例1比1或解析度640*480的圖片放在iPhone觀看，基本上會造成影像內容扭曲。 在影像處理研究領域中，解決此問題的技術為影像尺寸縮減。例如Scaling、Cropping和Warping。Scaling 和 Cropping兩者皆可以簡單且迅速地達成圖片壓縮，但前者會把影像中主要的物件壓縮，破壞圖片結構 而後者雖能依照使用者裁切出想要的區域，然而也會將其他重要物件摒棄。相較之下，Warping可以達到內容感知影像壓縮，但是仍具有時間消耗量大的問題。 Seam Carving是一個有效率且新穎的方法，藉由刪除或插入seam來完成影像縮減或擴增，達到內容感知影像壓縮。但此方法容易破壞影像中直線及物體邊緣的結構。因此，本篇論文著重實作Seam Carving技術，同時也強調影像中直線結構與物體邊緣的保護，來達成美觀的結果。此外，也將此研究延伸至影片壓縮，因為影格間的連續性，導致要處裡演算法更為複雜。實驗部分將會展示出壓縮後不同大小的影片及影像結果來驗證此論文所提出方法的可行和有效性。 With the various development of technology, there is getting more people use their digital or 3C products, which can enlarge and enrich our live. For instance, people used to take a photo with range finder camera but now it is popular for using digital camera or digital single lens reflex camera, even using smart phone. Unfortunately, it is hard to fit the image in different kinds of devices with various ratios. Take the iPhone with screen ratio for example, the height and width ratio is 3:2 people cannot put the image such as 1:1,640*480 those kinds of image into iPhone without distortion in visualization. The solution is to resize the image to the target ratio. There are several kinds of method to resize the image such as Scaling, Cropping, or Warping. Scaling and Cropping are the fast method to achieve such goal. But the previous one may distort the important object. Although cropping is the fast way to crop the interest area that the user wants, it may lose the other information too. Warping is another way to resize image, but it is not so effective as well. There is an effective and novel approach called seam carving that could achieve content-aware image resizing. Users may reduce or expand the height and width by themselves. But it may distort the line detection and object shape easily. In this thesis, we focus our research on improving the all line and shape structure in an image, trying to avoid seam goes through that kinds of things. Moreover, we discuss the seam carving on video resizing, which is more complicated than image resizing. At last, the experiments are carried out on different image/video frame with different number of seams. The results demonstrate the prominence of our method.
