隨著資訊系統的進步與網際網路的發達,多媒體系統的運用越來越多,影像資料庫又是相當重要的部分,如何從龐大的影像資料搜尋出使用者要的資料儼然成為一個熱門的議題。基於內容的相似影像尋取(Content-Based Image Retrieval,CBIR)是一查詢影像的重要方法。其中利用物件空間關係進行查詢是一常用的方法,然而在過往的研究中都沒有考慮到,影像中當物件之間發生重疊遮蔽時,其物件邊界不能確定的狀況,可能發生物件尋取上不夠完整的情形。因此本論文提出一種新的字串(2D B*-string),影像中當物件之間發生重疊遮蔽狀況時,能根據此字串中推論出可能之空間關係,進而進行影像推論及其相似性檢索。 When the image data becomes more and more, there are more applications of multimedia systems. Image database plays an important role. Hence, how to retrieve the desired image from the image database efficiently is a popular research topic. Content-Based Image Retrieval (CBIR) is an important approach to image query. Image retrieval based on spatial relationship between objects becomes popular. However, the situation in which objects overlap has not taken into account in previous research. Therefore, we presents a new representation called 2D B*-string in this paper. When objects overlap, the reasoning of relationship between objects can be performed based on the 2D B*-string.