CAJ | 학술논문

目的：为提高新疆维吾尔医草药图像的分类和检索准确率，对新疆维吾尔医草药图像进行感兴趣区域分割研究。方法分别采用阈值法、硬 C 均值（HCM）聚类算法和模糊 C 均值（FCM）聚类算法分割图像，将分割后的图像与原始图像进行迭代，分割15幅新疆维吾尔医药材图像，并将分割结果与手工分割结果进行比较，以评价分割算法的优劣。结果利用阈值法分割后的图像与手工分割图像进行比较，15幅图像的面积大小差异的平均值为12．7％，表面平均距离的平均值为2．79灰度值；利用 HCM 聚类算法分割后图像的面积大小差异的平均值为12．2％，表面平均距离的平均值为2．7灰度值；利用 FCM 聚类法分割后图像面积大小差异的平均值为9．04％，表面平均距离的平均值为0．96灰度值。结论阈值分割算法的分割速率比其他2种算法快，但该方法可能出现过度分割现象而导致信息的流失；硬 C 均值聚类算法能够较好地保留感兴趣区域，但分割时图像背景无法完全舍去而导致无用信息的掺杂；模糊 C 均值聚类算法分割速率相对较慢，但其整体分割准确率却高于以上2种算法，该算法较适于新疆维吾尔医药材图像的分割。本研究对后期图像的特征提取、分类和检索奠定基础。FCM method also could keep the interesting areas and also can wipe away more background area than HCM method．Therefore，FCM segmentation method was proved more appropriate for the Xinjiang Uygur medicine image segmentation，which served as a basic research on Xinjiang Uygur medicine image feature extraction，classification and retrieval．
목적：위제고신강유오이의초약도상적분류화검색준학솔，대신강유오이의초약도상진행감흥취구역분할연구。방법분별채용역치법、경 C 균치（HCM）취류산법화모호 C 균치（FCM）취류산법분할도상，장분할후적도상여원시도상진행질대，분할15폭신강유오이의약재도상，병장분할결과여수공분할결과진행비교，이평개분할산법적우렬。결과이용역치법분할후적도상여수공분할도상진행비교，15폭도상적면적대소차이적평균치위12．7％，표면평균거리적평균치위2．79회도치；이용 HCM 취류산법분할후도상적면적대소차이적평균치위12．2％，표면평균거리적평균치위2．7회도치；이용 FCM 취류법분할후도상면적대소차이적평균치위9．04％，표면평균거리적평균치위0．96회도치。결론역치분할산법적분할속솔비기타2충산법쾌，단해방법가능출현과도분할현상이도치신식적류실；경 C 균치취류산법능구교호지보류감흥취구역，단분할시도상배경무법완전사거이도치무용신식적참잡；모호 C 균치취류산법분할속솔상대교만，단기정체분할준학솔각고우이상2충산법，해산법교괄우신강유오이의약재도상적분할。본연구대후기도상적특정제취、분류화검색전정기출。FCM method also could keep the interesting areas and also can wipe away more background area than HCM method．Therefore，FCM segmentation method was proved more appropriate for the Xinjiang Uygur medicine image segmentation，which served as a basic research on Xinjiang Uygur medicine image feature extraction，classification and retrieval．
Objective To improve the accuracy of classification and retrieval of Xinjiang Uygur herbal medi-cine images by segmenting the interesting area from the image of the herbal medicine.Methods 15 Xin-jiang Uygur herbal medicine images were segmented by means of threshold method,hard C mean method and fuzzy C mean method.The whole medicine image was segmented by clustering similar pixels from dif-ferent pixel level by getting rid of useless group.The segmented image was iterated with the original im-age.The area size difference and the average surface distance were compared and evaluated.Results The average surface distance and the mean area size difference are 12.7% and 2.79 with threshold method, 12.2% and 2.7 with hard C mean method and 9.04% and 0.96 with fuzzy C mean method respectively.Con-clusion By three segmentation methods with 15 images,it's concluded that the threshold segmentation al-gorithm was proved simple and fast,but with some interesting areas missing and information missing.The HCM segmentation method could keep almost all the interesting areas but with some useless features.The FCM method also could keep the interesting areas and also can wipe away more background area thanHCM method.Therefore,FCM segmentation method was proved more appropriate for the Xinjiang Uygurmedicine image segmentation,which served as a basic research on Xinjiang Uygur medicine image featureextraction,classification and retrieval.