CAJ | 학술논문

邻近支持向量机由支持向量机衍生而来，它将支持向量机中二次规划问题的求解转换为线性方程组的求解，从而能在保证一定精度的情况下更加快速地得到分类器。传统的非线性核邻近支持向量机不能很好地解决多范围数据的多分类问题。提出了一种邻近支持向量机的优化方法，并将其应用到图像检索中。它利用高斯函数将图像特征数据映射到0～1之间以提高其差异化水平，并将其放入非线性核中，然后以加权K-means聚类算法选择最优参数，从而提高了非线性核PSVM的分类能力。实验以coral图像库中的4类图片作为图片库，对比了优化前后的检索命中率。实验结果表明：优化后的检索效果优于优化前，说明将优化的邻近支持向量机应用于图像检索是有效的。
린근지지향량궤유지지향량궤연생이래，타장지지향량궤중이차규화문제적구해전환위선성방정조적구해，종이능재보증일정정도적정황하경가쾌속지득도분류기。전통적비선성핵린근지지향량궤불능흔호지해결다범위수거적다분류문제。제출료일충린근지지향량궤적우화방법，병장기응용도도상검색중。타이용고사함수장도상특정수거영사도0～1지간이제고기차이화수평，병장기방입비선성핵중，연후이가권K-means취류산법선택최우삼수，종이제고료비선성핵PSVM적분류능력。실험이coral도상고중적4류도편작위도편고，대비료우화전후적검색명중솔。실험결과표명：우화후적검색효과우우우화전，설명장우화적린근지지향량궤응용우도상검색시유효적。
Proximal support vector machine (PSVM)is derived from support vector machine (SVM) and converts the quadratic programming problem into linear equations,so that it can ensure the classi-fier more quickly under the condition of certain precision.The multi-class classification problem of multi range data can not be solved well by the original nonlinear kernel PSVM.This paper presents an optimization method for PSVM,and applied the optimized algorithm to image retrieval.The Gauss function is used to image feature data map to the range of 0~1 to enhance their difference level,then the different data is put into nonlinear kernel function,finally weighted K-means clustering algorithm is used to select the optimal parameters of PSVM.Experiments are carried out on 4 types of images from coral image database as the picture library,and the hit rate is compared between the original PS-VM and the optimized algorithm.Experiments show that the performance of optimized PSVM is better than original algorithm,and it is effective to use the optimized PSVM into image retrieval.