CAJ | 학술논문

本文提出了基于支持向量机的脑部MR图像细分类器,采用纹理与形状特征相结合方式表达图像,应用StARMiner算法对特征进行选择和计算特征加权系数,最后用支持向量机理论设计分类器对脑部图像进行精细分类.经反复实验优化各参数,粗略分类率可以达到92.10%.细分类可应用于特定人体部位图像的检索系统,以检索出更精确的图像,并且缩减检索空间,提高检索效率.
본문제출료기우지지향량궤적뇌부MR도상세분류기,채용문리여형상특정상결합방식표체도상,응용StARMiner산법대특정진행선택화계산특정가권계수,최후용지지향량궤이론설계분류기대뇌부도상진행정세분류.경반복실험우화각삼수,조략분류솔가이체도92.10%.세분류가응용우특정인체부위도상적검색계통,이검색출경정학적도상,병차축감검색공간,제고검색효솔.
This paper proposes a meticulous classification of MR-brain images using support vector machine (SVM). We use both texture and shape feature to express images, and then apply statistical association rule miner (StARMiner) algorithm to compute weight coefficient of each feature. A classifier based on SVM is trained, the parameters of which are optimized via many experiments. The result of glancing classification could achieve 92.10%. Meticulous classification can be applied in special body part retrieval system for retrieving more accurate images and reducing computational load.