運用以內容為基礎之影像擷取於藥物辨識之研究
Applying Content Based Image Retrieval to Drug Identification Research
中文摘要
近年來知識水準的提升以及發生的一些重大藥物疏失事件，民眾對於藥物資訊 與藥物安全的需求與日劇增。可是要上網查詢這些藥物資訊，往往還是只能藉由 輸入文字的方式去查詢。基於資訊科技的進步與文字查詢的不便等各種因素，本 論文運用以內容為基礎的影像擷取方法，建置一個藥物實體影像辨識系統，提 供醫療人員或民眾運用數位相機等方式取得數位影像後，經由自動化的方式擷 取藥物影像的內容特徵進行分析，然後回饋使用者相關的藥物資訊。簡化了藥物 辨識系統的操作程序，增加民眾對藥物辨識系統的使用意願及方便性。
本論文探討影響藥物影像辨識的重要因素，並以GIFT 程式為範本針對藥物的
顏色、紋理、旋轉、遠近等因素進行驗證並提出本論文的作法與改進方式，在超 過七百張藥物影像的測試下，最後成功建立了一套藥物實體影像辨識系統，其 平均檢索成功率（Ｓ）可達80.11％，膠囊可達 94.92％。期望在未來可供藥物影 像辨識研究上的參考並能有助於藥物辨識系統新介面的研發，擺脫傳統文字查 詢的不便。
英文摘要
In recent years, with the improvement of knowledge level and some serious
medication errors happened, people require drug information and safety increasingly.
Mostly, we can only search drug information by typing some key words from
Internet. Based on the advancement of information technology and inconvenience of key word search, this thesis applied content-based image retrieval to build a Real Drug Image Identification System (RDIIS). This system allows health care personnel and the public upload digital images. After catching automatically the content of drug images, system will feed back users with relevant drug information. It simplifies the operation of drug identification system and increases the desire and convenience for the public to use it.
This thesis discussed many important factors which may influence drug image identification. We used GIFT as a model and tested many factors including color, texture, rotation and distance then proposed solution for improvements. In this study, we tested more than seven hundreds drug images and successfully build a RDIIS whose average of search success ratio (Ｓ) is able to get up to 80.11％ and capsule ratio is reached 94.92％. We expect that this thesis can be a reference for other drug image identification research and help the development of new user interface in order
to get rid of the inconvenience of the traditional key word searching.
