CBIR檢索效能改善策略之研究

Strategy Improvement for Content-based Image Retrieval

林孟鋒(Meng-Feng Lin)

指導教授 ： 李朱慧

朝陽科技大學/資訊學院/資訊科技研究所博士班/博士(2011年)

摘要

內容式影像檢索(CBIR)在多媒體資料庫中扮演一個重要角色，主要在於其系統能協助使用者檢索出相似的影像。一般而言，使用者在查詢時輸入範例影像，系統將就其低階特徵來協助進行檢索，所以如何運用CBIR來搜尋出使用者期望的結果便是一個重要的議題。因此，本論文設立不同的策略來協助系統改善影像檢索的效能，並且將多重查詢機制延伸應用於立體商標影像檢索。
首先，相關回饋機制是在檢索過程中收集使用者的檢索偏好，現行許多相關回饋研究主要是針對互動技術與執行準則進行探討。而本論文提出一個自我相關回饋策略，將藉由使用者回饋資訊來為資料庫中的每一張影像於相似度測量時設立屬於自己適合的結合權重，也就是給於每張影像屬於自我的相似度測量。
再者，多重查詢是一種較單一查詢更具有彈性和豐富的重要索引資訊的機制，協助使用者呈現更準確的索引意圖描述。而所謂的多重查詢是由使用者輸入多張的範例影像，其中每一張範例影像，都是呈現使用者的視覺感官與主要低階特徵的查詢代表，所以在索引處理過程中要如何整合多重查詢是一個重要的議題。並且就多重查詢特性可延伸應用於立體商標與立體影像的檢索，因此本論文主要在於發展有效率之多重查詢策略。其中，由於商標影像資料庫中商標註冊數量是迅速提升，並且註冊商標中文字說明與商標影像內容具有同等的重要性，加上一般二維商標附件只需一張影像，而立體商標是必須由數張影像構成三維成像，因此多重查詢為立體商標提供一個解決檢索的基礎。
綜整以上，本論文設立不同的CBIR改善檢索效能策略，並經由實驗證明其不同的策略分別有助於檢索相關影像的效能與提高使用者對於檢索結果的滿意度，相關敘述如下：第一，本論文將回饋記錄分別儲存於二個不同的儲存位置，如：基於索引影像(QRF-based) 或基於資料庫中被檢索出來的每一張相關影像(DBRF-based)；第二，本論文於CBIR下設立多個多重查詢策略，協助多重查詢處理，如：image-level與bin-level策略；最後，本論文將多重查詢機制延伸發展出一個適用於立體商標的檢索系統。

關鍵字

內容式影像檢索 ； 多重查詢 ； 索引處理 ； 相似度測量 ； 相關回饋 ； 立體商標

並列摘要

Content-Based Image Retrieval (CBIR) leads a central role in the multimedia database. The system can retrieve the similar images through user’s query information. Each query is represented by low-level features of image. However, how to retrieve users expected result in CBIR is an important issue. This dissertation proposed difference strategies to improve the image retrieval performance. Besides, multi-query is extended to apply for three-dimensional trademark retrieval.
Relevance feedback mechanism in the retrieval phase of CBIR collects the user retrieval preference. Many researches of relevance feedback focus on interactive techniques and implement criteria. This dissertation proposed an ego strategy of relevance feedback which can set adaptive weights of similarity measurement for each image in the database from user’s feedback, i.e. ego-similarity measurement.
Multi-query images have support flexibility and plenty important information than single-query image. Multi-query contains several query images selected by the user and each image in multi-query has its visual representation in user’s respect and major low-level features. Therefore, how to integrate multi-query in the query processing is an important issue. There are several interesting applications for multi-query operations, such as three-dimensional trademark and stereo image. The efficient multi-query strategies are developed in this dissertation.
An application of multi-query, the number of registered trademarks in the image trademark database has risen rapidly. Both the text description of the trademark registration and the image content of the trademark are important. One three-dimensional trademark consists of several images, whereas a two-dimensional trademark contains only one. Multi-query seems to provide a solution for three-dimensional trademark retrieval.
Experimental results manifest that the proposed strategies strengthen the performance of the retrieved relevant images and let the result be satisfied by the user. First, this dissertation would explore the feedback records archived in the two different ways that stored along with query image (QRF-based) or along with each retrieved relevant image from the image database (DBRF-based). Second, this dissertation proposed multi-query strategies which were provided to help the query processing in CBIR, including image-level strategy and bin-level strategies. Finally, multi-query approach is applied to develop a retrieval system for three-dimensional trademarks, to integrate the similarity of each image in one set.

並列關鍵字

Multi-query ； Query processing ； Relevance feedback ； Three-dimensional trademark ； Similarity measurement ； Content-based image retrieval

參考文獻

[2]A. Amato and V.D. Lecce (2008), “A knowledge based approach for a fast image retrieval system,” Image and Vision Computing, Vol. 26, No. 11, pp. 1466-1480.

[3]M. Arevalillo Herraez, F.J. Ferri and J. Domingo (2010), “A naive relevance feedback model for content-based image retrieval using multiple similarity measures,” Pattern Recognition, Vol. 43, No. 3, pp. 619-629.

[4]M. Arevalillo Herraez, F.J. Ferri and S. Moreno Picot (2011), “Distance-based relevance feedback using a hybrid interactive genetic algorithm for image retrieval,” Applied Soft Computing, Vol. 11, No. 2, pp.1782-1791.

[5]M. Arevalillo Herraez, M. Zacares, X. Benavent and E. De Ves (2008), “A relevance feedback CBIR algorithm based on fuzzy sets,” Signal Processing: Image Communication, Vol. 23, No. 7, pp. 490-504.

[6]E. Baratis, E.G.M. Petrakis and E. Milios (2008), “Automatic website summarization by image content: A case study with logo and trademark images,” IEEE Transactions on Knowledge and Data Engineering, Vol. 20, No. 9, pp. 1195 -1204.