圖像資料庫系統:以形狀或顏色作為查詢依據

Abstract

隨著資料庫容量的增大, 傳統式的查詢方式 -- 以文字作為查詢依據 , 已逐漸不能滿足使用者的需求. 有鑒於此, 以文圖形內容作為查詢的方式 便成為研究人員努力的目標. 本論文針對圖像資料庫系統提出一種新的描 述子, 用以作為內容導向式查詢的依據. 為了證明此形狀描述子的有效 性, 我們實作了一個圖像資料庫系統; 同時, 為了提高系統的準確度, 我 們選擇另一種傳統的描述子來與新描述子做組合式的查詢. 在一般的圖像 資料庫系統中所使用的形狀描述子大多需要高維度的搜尋空間才能獲得令 人滿意的有效性, 而本系統中所使用的描述子具有快速及簡易的特性, 在 需要即時交談的圖像資料庫系統中是相當合適的. 另外, 我們也探討了一 些有助於內容導向式查詢的技術, 以期在未來的實際應用上能獲致更佳的 成果. Because the size of a pictorial database becomes larger gradually, users cannot be satisfied with the traditional search method -- query by text. To solve this problem, many researchers are engaged in the development of content-based searches. In this thesis, we propose a new descriptor as the basis of content-based searches. To prove its effectiveness, we implement a pictorial database system using this descriptor. In this system, we also choose another shape descriptor and combine it with our proposal to perform queries. The experimental results are evaluated by the effectiveness. To acquire a good effectiveness, high dimensionality of search space is necessary for most of pictorial database systems. Our descriptor can be calculated without complex computation, so it is appropriate for the real-time database system. Besides, we also present some mechanisms that can benefit the content- based searches and can apply very well to many pictorial database systems.