區域相鄰關係為基礎的影像搜尋法

Image Retrieval by Using Region Adjacency Relations

黃育仁(Yu-Len Huang) ； 李建平(Jian-Ping Lee) ； 陳冠良(Guan-Liang Chen) ； 李志忠(Zhi-Zhong Lee) ； 張美萍(Mei-Ping Chang) ； 黃秀菊(Xiu-Ju Huang)

《東海科學》 6期 (2004/07) Pp. 17-41

https://doi.org/10.29723/TS.200407.0002

摘要

隨著網際網路蓬勃之發展，網路資料不再侷限於文字部份，多媒體資訊在網際網路上應用更形重要，其中又以影像資料的使用率最高。如何有效的在網際網路上搜尋所需要的多媒體資料，已成為當前的研究重點。本論文提出以區域相鄰圖(region adjacency graph, RAG)為基礎之影像搜尋系統，利用影像中各區域的關係來搜尋相似的影像，並結合特徵擷取、影像切割的技術，使影像搜尋不受旋轉、放大、縮小以及部份子圖的限制，使得搜尋出之影像合乎使用者的要求。在實際應用於商標影像之搜尋實驗，此系統得到了相當良好的效果，更印證了此影像搜尋法之可行性。本研究最主要的目的在於讓網際網路的搜尋侷限於文字，還可擴展至影像及多媒體資料。

關鍵字

網際網路 ； 區塊相鄰圖 ； 影像搜尋 ； 商標 ； 多媒體資料庫

並列摘要

With the growth of multimedia application and the spread of Internet, the access of digital image becomes effortless. Hence, the image retrieval is essential for multimedia libraries and databases. Although manual image annotations can provide a certain of help for retrieving images, it was still a complicated question to large collections of digital images. There is a new focus on computer automated approaches for image retrieval. This paper proposes the use of the region adjacency graph (RAG) to query the similar images. Each image in the image database was first segmented into meaningful regions. Features (color and shape) and spatial relations of regions were performed to retrieve images in the image database. The proposed method diminishes the influences upon image rotation, zooming, cropping for image retrieval. Computer simulation results show that the proposed method achieves a good retrieval performance in a trademark image database. From the experimental results, it shows that the proposed method is expected to be very useful for image retrieval. In future work, we will extend the proposed method to natural image and multimedia database applications.

並列關鍵字

Internet ； region adjacency graph ； image retrieval ； trademark ； multimedia database