图像识别一般是一个高维分类问题,不可避免会存在很多冗余属性。针对图像识别问题,提出一种基于支持向量机预分类的属性选择算法。该算法首先在原始训练集合上利用支持向量机算法进行预分类,求取分类超平面。然后根据分类超平面的方向量的系数大小,进行特征属性选择。在Corel Image数据库上的仿真实验表明该算法是一种优秀的属性选择算法,可以有效地提高传统分类算法的分类性能。%Image recognition is generally a problem of high-dimensional classification , so usually it inevitably contains a lot of redundant attributes.Aiming at image recognition problem , in this paper we propose an attribute selection algorithm which is based on SVM pre -classification.First, we use SVM on original training set to pre-classify for computing the classified hyperplane .Then we select features attribute according to the size of the coefficient of the classified hyperplane .Simulation experiments made on the Corel Image data set show that the proposed algorithm is an effective attributes selection method , it can effectively improve the classification performance of traditional classification algorithm .
展开▼