影像資料庫中查詢影像物件之特徵擷取

Abstract

本文針對影像資料庫的影像查詢提出一套擷取的方法,與一般傳統查詢不 同的是,我們是以影像物件的特徵當作索引來擷取影像資料庫的資料,這套 以影像處理為基礎的系統提供使用者擷取出具有某個影像物件的影像.在 此提到的影像物件的特徵包括了形狀與色彩,而有效的擷取通常取決於影 像特徵的選取與計算,為此我們配合使用者的查詢,結合形狀與色彩成為一 種影像特徵來作影像資料庫的擷取.本系統影像資料庫的設計特點是以物 件為導向索引,配合影像辨識碼的串列而形成.物件比對的索引是基於影像 物件的特徵,當目的物件擷取出來後,包含有這些物件的影像將從串列中一 一被擷取出. In this thesis we have proposed access methods for the retrieval of images in image databases. Differing from conventional image database querying, image object contents are used to query image databases. The image content retrieval system provides users the retrieval capability for querying images of desired objects by using image processing techniques. Features of image objects described in this thesis include color features and shape features. Key issues in our system include the derivation and computation of image features that provide useful retrieval functionality. We use a merged feature , including shape and color features, to access image database by a query method. The speciality of the design of our image database is object oriented. All images in databases are linked to special objects that occur in these linking images. The matching indices are based on the feature of objects. When the target object is accessed by an input object, the images including the target objects are retrived through the link associated with the object.