CAJ | 학술논문

目的：利用决策树分类方法探讨一种新的图像特征优化降维方法。方法首先利用图像滤波、灰度阈值、腐蚀运算等图像操作对新疆草药图像做分割预处理，获取草药图像感兴趣区形状，然后提取草药形状的 Fou-rier-Mellin 矩、Hu 矩等形状特征及图像主颜色直方图特征共14个特征分量；最后应用决策树 C4．5分类算法，在不同草药形状特征分量组合下比较草药图像分类准确率，经过优化分析，选用 Hu 矩的 H2、H4和图像主颜色直方图特征的 C1～C3特征分量构成图像形状颜色混合特征的5个特征分量，用于草药图像分类。结果决策树分类方法准确率达到88．55％；将优化后的图像形状颜色混合特征用于图像检索，其检索结果集前60张图的平均查准率达到89．31％，实现形状特征分量的优化降维。结论利用决策树分类方法可以进行图像特征优化降维，降低图像形状颜色混合特征维数，提高图像分类准确率和图像检索效率，为进一步图像特征的优化研究提供参考。
목적：이용결책수분류방법탐토일충신적도상특정우화강유방법。방법수선이용도상려파、회도역치、부식운산등도상조작대신강초약도상주분할예처리，획취초약도상감흥취구형상，연후제취초약형상적 Fou-rier-Mellin 구、Hu 구등형상특정급도상주안색직방도특정공14개특정분량；최후응용결책수 C4．5분류산법，재불동초약형상특정분량조합하비교초약도상분류준학솔，경과우화분석，선용 Hu 구적 H2、H4화도상주안색직방도특정적 C1～C3특정분량구성도상형상안색혼합특정적5개특정분량，용우초약도상분류。결과결책수분류방법준학솔체도88．55％；장우화후적도상형상안색혼합특정용우도상검색，기검색결과집전60장도적평균사준솔체도89．31％，실현형상특정분량적우화강유。결론이용결책수분류방법가이진행도상특정우화강유，강저도상형상안색혼합특정유수，제고도상분류준학솔화도상검색효솔，위진일보도상특정적우화연구제공삼고。
Objective To study an innovative method of dimension reduction of image feature by means of C4.5 decision tree classification algorithm.Methods Firstly,such operations as image filtering,gray threshold and corrosion arithmetic were applied to segment Xinjiang herb image for obtaining the region of interest (ROI)of herb shape.Secondly,14 feature components,including herbal image shape Fourier-Mellin moments, Hu moments and the herbal image main color histogram features were extracted. Finally,C4.5 decision tree classification algorithm was adopted and 5 optimized components were identified as the image mixed shape-color features for the herbal image classification.Results The classification ac-curacy rate of this method was proved to be 88.55%;the average precision rate in the top 60 images re-trieval with the optimized mixed shape-color feature was up to 89.31%.Conclusion The optimization and dimension reduction of image feature could be realized by means of C4.5 decision tree classification algo-rithm to provide a reference for the further optimization of image features.