使用乏晰空間關係表示法的影像擷取

Abstract

以內容為基礎之影像擷取使用影像的色彩、紋理、物件的形狀以及物件之間的空間關係，來作為擷取類似影像的依據。在本論文中我們專注於空間關係的討論，並且提出一個結合了圖像索引法則和乏晰邏輯的影像表示法，可以將物件之間不明確的空間關係評估出來。此一特性可應用於地理資訊系統中的空間關係推理。同時它的乏晰性質和人類對於空間關係的認知較為相近，使得此一影像表示法可以很容易地和自然語言結合，有利於在空間資料庫中進行自然語言查詢。

Content-based image retrieval retrieves similar images via the contents of the query image, such as color, texture, shapes of objects, and the spatial relationships between objects. In this thesis, we focus on the spatial relationships of the content-based image retrieval and propose a new representation which combines iconic indexing scheme and fuzzy logic to provide the ability to assess the ambiguous spatial relationships between objects. This representation captures the ambiguity of spatial relationships and therefore facilitates spatial reasoning which is helpful in the area of geographic information systems (GIS). And the fuzzy property of the representation also makes it more consistent with human concepts and easier to incorporate with natural language, and therefore facilitates natural language query in the spatial databases.