結合外型和空間關係之影像擷取研究

Abstract

在本篇論文裡，我們提出一個以內容為主的影像擷取系統，利用鏈碼技術來表示影像的外型特徵。再將鏈碼轉換成為三鏈碼以節省空間，而三鏈碼則是由三個有序鏈碼所構成。我們還提出三鏈碼的相似度測量公式，以擷取外型相似的影像。本方法可以處理單物件影像及多物件影像。對於單物件影像查詢，查詢影像與資料庫中所有影像做外型相似測量。對於多物件影像查詢，查詢影像先與資料庫影像做空間關係比對，以過濾不相似影像。最後我們以實際影像資料驗證本系統對於縮小、旋轉或手繪影像查詢，都能擷取出相似的影像出來。

In this paper, we propose a content-based image retrieval system which uses a chain code technique to represent shape feature. We transform chain code into tri-chain code to reduce storage space. Tri-chain code is constructed by combining every three ordered chain code. Our propose method support single-object images query and multi-object images query. For single-object image query, we use query image’s shape descriptor and compare it with all shape descriptors stored in the image database. For multi-object query, we propose a spatial relationship matching to filter non-similar images in image database. We have illustrated that the system can retrieve the similar images from database by using scaled, rotated , or sketched query images.