以內容導向擷取技術產生影像馬賽克之研究

本研究提出一個以內容導向影像擷取 (content-based image retrieval, CBIR) 產生影像馬賽克 (image mosaic) 之方法，主要分為三個步驟：(1) 原始影像切割、(2) 影像資料庫比對、(3) 影像鑲嵌重組。第一步驟係指將原始影像切割成子影像 (tile image)；第二步驟以內容導向影像擷取為主，著重影像特徵值之相似度比對；第三步驟以最相似影像取代並鑲嵌重組成一張遠看為原始影像，近看卻包含多種相異影像的影像，稱為馬賽克影像。本研究將影像特徵值分成色彩與紋理兩部份。色彩特徵值以 MPEG-7 之色彩描述子 (color descriptor) 為依據，將三原色轉成 HSV 值後，再轉成 HMMD 值；紋理特徵值以 MPEG-7之紋理描述子 (texture descriptor) 為依據，結合檢索紋理標準之Gabor 函式與快速傅立葉轉換 (fast Fourier transform)。此外，本研究運用自我組織映射 (self-organizing map) 將影像資料庫中的影像分類，並建立索引結構，以減少比對次數，進而提昇搜尋效率。最後，透過歐基里德 (Euclidean) 公式取得最相似影像，取代原始影像後鑲嵌重組，即產生馬賽克影像。本研究設計兩種績效評估之方法：(1) 主觀評估：係指與現有軟體影像蒙太奇 (PhotoMontage) 比較作品與原始影像之相似程度，且採取比例逐次放大之方式，判斷受測者由遠至近觀測馬賽克作品所產生的變化；(2) 客觀評估：係指透過 PSNR 值計算本研究作品於比例逐次放大之情況下，與原始影像間之關係。

關鍵字

影像馬賽克；特徵萃取；自我組織映射；內容導向影像擷取

並列摘要

Image mosaic is a technique by which many small tile images are used to tessellate a large image. The large image, called a mosaic image, can convey its own visual contents as a whole, while each of its tile images also has its meaningful contents. The processing of image mosaic consists of three major steps: (1) image partition, (2) image retrieval, and (3) image tessellation. First, an input image is partitioned into many blocks, from each of which texture and color features are extracted. Second, for each image block, we calculate its similarity with each database image and then retrieve a best similar database image. Finally, each image block is replaced by its best similar database image. According to MPEG-7, we calculate HMMD values from each tile image, followed by using the Gabor function and Fast Fourier Transform to extract texture features. Then we classify all database images by self-organizing map to improve the retrieval efficiency. In this study, we propose two approaches to evaluating image mosaic’s quality. The two approaches are based on the subjective perception of humans and PSNR (an objective evaluation). As for the subjectivity test, we design a questionnaire and analyze the response of users by enlarging the original images gradually. As for the objectivity test, we calculate PSNR between the mosaic image and the original image. According to our analysis, we can find a relationship between the two tests.

並列關鍵字

Image mosaic ； feature extraction ； self-organizing map clustering ； content-based image retrieval

參考文獻

[3] 陳品秀譯 (2005)：馬賽克創作技法小百科，第一版，城邦文化事業股份有限公司，台北市。

[46] Y. Zijun and C. C. Jay Kuo (1999), “Survey on image content analysis, indexing, and retrieval techniques and status report of MPEG-7,” Journal of Science and Engineering, Vol. 2, No. 3, pp. 101-118.

[10] A. Yoshitaka and T. Ichikawa (1999), “A survey on content-based retrieval for multimedia databases,” IEEE Transactions on Knowledge and Data Engineering, Vol. 11, No. 1, pp. 81-93.

[11] A. W. M. Smeulders, M. Worring, S. Santini, A. Gupta, and R. Jain (2000), “Content-based image retrieval at the end of the early years,” IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol. 22, No. 12, pp. 1349-1380.

[12] C. S. Fuh, S. W. Cho, and K. Essig (2000/1), “Hierarchical color image region segmentation for content-based image retrieval system,” IEEE Transactions on Image Processing, Vol. 9, No. 1, pp. 156-162, New York, NY, USA.

國際替代計量

以內容導向擷取技術產生影像馬賽克之研究

全文下載

主題瀏覽