摘要
有些时候单独用常用的线性、多项式、Gauss/RBF径向基和SIGMOID核函数构造支持向量机(Support Vector Ma-chine,SVM)进行图像分割,并不能得到满意的结果。为了得到更好的分割效果,文中提出一种基于最优权值组合核函数的支持向量机图像分割方法,将作为局部核的Gauss/RBF核函数、全局核的多项式核函数,以及广泛运用的SIGMOID核函数通过两两加权来构造新的函数,并对权值进行遍历优化,找出分割效果最好的权值组合。实验结果表明,多项式核函数和SIGMOID核函数加权形成的核函数的分割效果最好,并且不同的权值对该组合核函数的分割效果影响很小,权值选择有更大的自由度,可以作为进一步研究核函数的基础。
Sometimes support vector machines （SVMs） formed by commonly used kernel function, such as linear, polynomial, Gauss RBF and SIGMOID kernel function, segmenting image can not obtain the satisfactory results. In order to get better segmentation result, put forward a method of image segmentation based on SVM composed by the optimal weighted combination kernel functions. Put it another way, construct a new kernel function by weighting each two of polynomial kernel function regarded as local kernel function, Gauss RBF kernel function regarded as global kernel function and SIGMOID kernel function which is widely used. Besides, optimize the weights to find the best weights for the segmentation effect. The experimental results show the combination of polynomial and SIGMOID kernel function proposed has very good and stable effect for SVM segmentation, which can be the foundation of the following up study of kernel function.
出处
《计算机技术与发展》
2013年第3期96-100,共5页
Computer Technology and Development
基金
国家自然科学基金资助项目(61070234
61071167)
关键词
图像分割
支持向量机
核函数
加权组合
遍历优化
image segmentation
support vector machine
kernel function
weighted combination
optimization of traversal
