網際網路影像搜尋引擎之設計與建構

The Design and Implementation of an Image-based Search Engine

周佰瑜(Pai-Yu Chou)

指導教授 ： 李錫捷

元智大學/資訊學院/資訊管理學系/碩士(2003年)

摘要

一般的影像檢索資料庫在作法上大多是採用人工先對圖案加以標註文字，再對文字做索引的方式來檢索影像，這樣的作法無法客觀的描述影像本身的特性，本研究提出了一個以顏色分布、外型及材質為特徵的的檢索法方法，可以客觀的描述影像本身的特性及關連，使用者可用此系統作為分類的依據或是輔助的系統。
此系統運用原有文字全文檢索的技術加上影像的特徵擷取，發展了影像及文字二用的檢索技術，使用這種架構可以把新的影像檢索方法架構在原有成熟的技術上。
以往的影像檢索資料庫大多需耗費許多人工來維護，本系統由資料取得到特徵擷取皆不需經由人工處理，可有效的省去人工維護的困擾。本系統使用的特徵為影像的週邊文字、顏色分布、外型及材質，經評比有73%的準確度並結合平行架構以有效的增進處理的效能。

關鍵字

全文檢索 ； 搜尋引擎 ； 影像檢索引擎 ； 反轉索引檔 ； 材質 ； 外型 ； 顏色分布

並列摘要

The most common method used by the current available image search engines is first to manually annotate images using texts, and then perform the search using traditional text indexing methods. One of the problems with this technique is the inability to describe objectively various attributes of an image. In this study, we propose an image indexing method that uses histogram, shape, and textures of an image to build an objective description of a given image. This approach can also be used along with an existing classification system.
The proposed system integrates the existing text-based retrieval techniques and the image attributes acquisition mechanisms. As a result, it allows the application of new image indexing methods onto existing search technologies.
Most of currently existing image index databases require maintenance with considerable human resources. Since the proposed system finds histogram, shape, and textures as attributes of an image automatically, the need for human resources is minimized. Experiments are conducted and the benchmark results have demonstrated a classification rate of 73%. To improve the performance of the proposed system, parallel algorithms are also applied for indexing.

並列關鍵字

Information Retrieval ； Search Engine ； Image-based Search Engine ； Inverted File ； texture ； shape ； histogram

參考文獻

C.-H. Lee and H.-C. Yang, "Developing an adaptive search engine for e-commerce using a web mining approach," in Proc. Int. Conf. Inform. Technol.: Coding and Computing, 2001, pp. 604--608.

S. Brin and L. Page “The anatomy of a large scale hypertextual web search engine. “ In Proc. of 7th World-Wide Web conference (WWW7), (1998)

Google Inc.

M.R. Henzinger. "Hyperlink analysis for the web." IEEE Internet Computing, 1:45--50, 2001.

J. Thomas, R. Watson “Decoding image semantics using composite region templates”, Content-Based Access of Image and Video Libraries, 1998. Proceedings.