- 學位論文
區域性影像檢索架構之研究
The Framework Study of Region-Based Image Retrieval System
摘要
內涵式之影像檢索(Content-Based Image Retrieval)已發展許久，許多文獻提出以色彩、紋理和形狀等資訊為影像內容之特徵，但如何整合這些特徵資訊以進行分類及檢索，實為一艱困之問題。 本研究提出一套區域性影像檢索(Region-Based Image Retrieval)之架構，對影像進行區塊化及物件化，可彈性將色彩、紋理和形狀等特徵資訊加入區塊物件中，對於影像檢索方面，可使用物件化後影像區塊之樹鏈物件進行比對，比對之法則可彈性改變，適合搭配各種理論進行比對，如模糊理論(Fuzzy Theory)或灰色理論(Gray Theory)等，而本研究採用模糊推論。 最後，為驗證本研究所提出之方法，利用以圖找圖之方法進行影像檢索，測試時分為封閉式測試及開放式測試，採用之影像特徵資訊有區塊色彩及區塊尺寸大小，而區塊物件比對法則利用模糊理論進行比對。經由實驗結果證明，本研究所提出之區域性影像檢索架構確實為可行。
並列摘要
Several content-based image retrieval (CBIR) systems have been developed and utilized to provide effective retrieval of image data based on their content. Furthermore, region-based image retrieval (RBIR) has been adopted, according to the concept of object by human vision, to improve the performance in content-based image retrieval. This thesis proposes an efficient framework in RBIR which utilize not only the features of each region but also the correlations between the neighboring regions. A tree-based indexing scheme is applied to account the retrieved result by the index of tree-link compare value (TCV) based on the region adjacency graph (RAG) with the corresponding region similar value (RSV) and node ration value (NRV) at the different nodes. For the RSV computation, different image features, such as color, size and texture could be adopted into account. In this thesis, both the color and size information of neighboring regions are considered into RSV computation. Then, the fuzzy reasoning mechanism is applied to calculate the region similarity of color information and RSV. Finally, the experimental results are verified the performance of this proposed image retrieval system. Moreover, both the opening-type and closing-type image database are used to test the robustness and reliability of the proposed system. The results show that this algorithm can accurately retrieve the query image from the image database.