利用色彩與空間資訊之高效率影像檢索系統

An efficient color indexing system for image retrieval using color and spatial information

呂明祐(Ming-Yu Lu)

指導教授 ： 陳永盛

元智大學/工程學院/電機工程學系/碩士(2002年)

摘要

色彩在包含影像內容為基底的影像檢索系統中是相當重要的特徵，因為色彩特徵只須建立直方圖即可很容易的被擷取出來，而且適合做相似度的比較與計算。但只以色彩特徵為影像資料檢索時，卻有一個嚴重的缺點，即缺少了色彩特徵所在位置之相對應空間資訊，本論文提出一套以色彩特徵並包含空間資訊的影像檢索系統。此影像檢索系統，由影像資料庫建立與影像資料庫檢索兩個子系統所組成，在建立影像資料庫系統時，將RGB色彩空間轉換成HSV色彩空間，再以此色彩空間進行包含空間資訊之色彩特徵擷取，然後建立影像資料庫。此影像資料檢索系統，我們提供多重色彩及空間檢索條件之輸入、多種常用相似度比較方法的選擇以及利用上述條件融合不同影像而產生新的影像。實驗結果證實我們所提出方法的可行性。

關鍵字

影像檢索 ； 色彩擷取 ； 影像資料庫 ； 相似度計算 ； 色彩直方圖

並列摘要

Color is one of the most prominent perceptual features. Most commercial CBIR systems include color as one of the features. The easy-to-compute color histogram is a popular and widely used image feature for image retrieval. However, when doing color features on the image database retrieval, there is a serious weak point that is lack of spatial information for the color feature. Due to this reason, in this thesis we will present a set of color feature including spatial information for the construction of our image retrieval system. This image retrieval system consists of image database indexing and image database retrieval subsystems. During the establishment of image database system, it converts RGB color-space into HSV color-space. Then utilize the color space to extract color feature that is including in space information, which is able to establish the image database. In the Image Database Retrieval System, we provide multi-color, multi-region retrieval conditional inputs, and several common-used similarity-computation selections, and new image produced with color-spatial retrieval results. Experiments confirm the feasibility of the implemented system.

並列關鍵字

Histogram ； Median Filter ； Similarity ； Image Indexing ； Image Retrieval

參考文獻

[1]G. Salton. Automatic Text Processing─the Transformation, Analysis and Retrieval of Information by Computer, Addison-Wesley Publishing Co., Reading, MA, 1989.

[4]H. Schweitzer, Template matching approach to content based image indexing by low dimensional Euclidean embedding, IEEE Proc. of International Conference on Computer Vision, Vol. 2, pp.566 -571, 2001.

[5]R. Distasi, M. Nappi, M. Tucci, S. Vitulano, Context: a technique for image retrieval integrating Contour and Texture information, IEEE Proc. of International Conference on Image Analysis and Processing, pp.224 —229, 2001.

[6]R. Pirrone, M. L. Cascia, Texture classification for content-based image retrieval, IEEE Proc. of International Conference on Image Analysis and Processing, pp.398-403, 2001.

[7]G. Voulgaris, J. Jiang, Texture-based image retrieval in wavelets compressed domain, IEEE Proc. of International Conference on Image Processing Vol.2, pp.125-128 , 2001.