基于子空间选择的SVM快速算法及其在医学图像奇异点检测中的应用

支持向量机（SVM）花费大量时间用于对非支持向量样本的优化．根据支持向量都是位于两类边界的靠近分类超平面的样本点，本文提出首先利用基于中心距离比值法排除大部分远离分类超平面的样本，然后以最小类间距离样本数作为测度进一步选择边界样本．得到包含所有支持向量的最小样本集，构成新的训练样本集训练SVM．将提出的算法应用于解决医学图像奇异点检测问题．实验结果表明．该算法减小了训练样本集的规模，有效地缩短了SVM训练算法的时间．同时获得了较高的榆出率．

A Fast SVM Algorithm Based on Subspace Selection of Samples and Its Application in Outliers Detection in Medical Images

Support vector machine(SVM) takes huge time for optimization of samples of unsupported vectors.And this makes it time-consuming to train supported vector machine classifier.The supported vectors are located near the region of hyper-plane of two classes.Based on ratio of distance between the centers of two classes,a cropping method to get a smaller training set was first represented.Moreover,the least distance between two classes was considered and the sample set was cropped.The sample set selected at last not only was smallest,but also covered all supported vectors.The proposed algorithm was used to solve the detection problem of image bizarre points.The results of experiments showed that the size of training set and abbreviates time of training stage could be reduced efficiently.At the same time,it gains a higher detection rate than the method based on SMO was achieved.