基於多重影像特徵與權重調整的口服藥物影像辨識之研究

The Study of Automatic Drug Images Identification System Based on Multiple Image Features and Adjust Features Weights

陳英豪(Ying-hao Chen)

指導教授 ： 陳榮靜

朝陽科技大學/資訊學院/資訊管理系/碩士(2010年)

摘要

近年來，由於用藥疏失的案例發生，使得民眾對於用藥安全的相關知識需求逐漸增加。口服藥物可以根據不同成份、含量、用途分為許多種類。但民眾並不是醫療人員具有專業的藥物資訊能夠辨識藥物。因此只能利用藥物名稱與描述藥物外觀的特徵等關鍵字，藉此搜尋藥物的相關資訊。目前各大醫療機構如區域型醫院，大多都有提供關鍵字搜尋的藥物辨識系統。然而，人們在描述藥物特徵都是依照各自的觀點來描述藥物特徵，如彩色、形狀的描述等。這些藥物的描述並無一致性說法，使得系統無法成功辨識進而造成錯誤。本研究介紹現有的藥物辨識系統，並提出一套以內容為基礎的影像擷取技術，透過取得口服藥物影像中的色彩、形狀、比例、大小、紋理特徵。利用權重歐基理德距離公式計算藥物的相似度測量，且根據模糊規則調整權重，建立一個自動化藥物影像辨識系統(Automatic Drug Image Identification System, ADIIS)提供藥物辨識。使用者只需拍攝藥物的數位影像，系統便可以自動辨識藥物。為能夠完善的提供藥物資訊與避免辨識的錯誤，本系統列舉出前十名辨識率最高的藥物做為結果，輔助使用者有效的辨識藥物。經由實驗證明，在第一名的正確率能達到92.6%，且在順序七以前皆能成功辨識，證明本研究在於口服藥物辨識是有效且具可行性。

關鍵字

Hamming距離 ； Gabor濾波 ； CBIR ； 權重相似度 ； 類神經網路

並列摘要

In order to take medicines safely, it’s better for a patient to know drugs information first. Drugs can be categorized into many kinds, such as different compositions, content and shapes. However, users do not always possess or comprehend professional drug facts. Many drug recognition systems offer keyword searches, but are difficult for users to understand the medications’ names. One feasible way would be letting users describe the features of drugs according to their appearance, such as color, shape, etc. In this paper, we propose an automatic drug image identification system (ADIIS) based on multiple image features. ADIIS is able to improve drug identification errors as well as provide more accurate drug information. In our primary experiments, by using an image, the system is able to retrieve the top ten similar drugs for the user to identify the specific one. Experiment results show that the first one of the ten drugs identified by ADIIS up to 92.6% correct rates. In addition, the system can successfully identify drug image in the top 7 of outcome.

並列關鍵字

Weight Similarity ； Neural Network ； CBIR ； Hamming distance ； Gabor filter

參考文獻

[3] W. Chen, P. J. Chao, and H. L. Lin (2007), “Drug identification by network adaptive content-based image retrieval,” The Journal of Health Science, Vol. 9, No. 2, pp. 133-145.

[1] A. Abdullah, R. C. Veltkamp, and M. A. Wiering (2010), “Fixed partitioning and salient points with MPEG-7 cluster correlograms for image categorization,” Pattern Recognition, Vol. 43, No. 3, pp. 650-662.

[2] S. Arivazhagan, L. Ganesan, and S. P. Priyal (2006), “Texture classification using Gabor wavelets based rotation invariant features,” Pattern Recognition Letters, Vol. 27, No. 16, pp. 1976-1982.

[4] J. X. Du, D. S. Huang, X. F. Wang, and X. G (2007), “Shape recognition based on neural networks trained by differential evolution algorithm,” Neurocomputing, Vol. 70, No. 4-6, pp. 896-903.

[7] Z. Geradts, H. Hardy, A. Poorman, and J. Bijhold (2001), “Evaluation of contents based image retrieval methods for a database of logos on drug tablets,” Image Analysis and Characterization of SPIE, Vol. 4232, pp. 553-562.