影音媒體之建築物圖像偵測

Abstract

從圖像本身的內容偵測出兩張圖像間是否含有相同的建築物是個困難且極富挑戰性的工作。如果能夠藉由圖像內容進行檢索，就可避免使用關鍵字檢索引起的各種問題。本篇論文中提出一個以紋理特徵為基礎的建築物匹配方式。首先從圖像中截取出特徵，並找出兩張圖像之間相同的特徵。然後，將結果中不合理的部分刪除並使用此結果找出兩張圖像中是否含有相同建築物。因為並未使用建築物模型，所以任何類型的建築物偵測皆可使用，不需做其他調整。我們以4支介紹景點的影片中所擷取出來的圖像及16張在影片中出現的建築物之圖像，驗證所提的方法能正確的找出圖像中相同的建築物。論文最後更進一步將所提的方法整合至一個已有的影音網站中並成功應用在影片中景點的定位上。

Detecting the same building between two images by their content is an important and challenging task. If we can query images by the content getting from an image, then we avoid all problems which keyword-search causes. In this paper, we propose a method for building mapping based on texture feature. First, we get the features from two images and find same features between two images. Then, we discard some of the results in unreasonable and use the result to find out whether the two images contain the same building. Because we don’t use any building models, so the algorithm can detect any type of building and does not need to adjust model for different buildings. The efficacy of the proposed method is tested on the images getting from 4 videos and 16 building images which are appeared in the videos. Moreover, we integrate the proposed method to an existed video web site and locating the attractions in the video successfully.