摘要
为了提高图像分割的质量,采用支持向量机核函数算法。首先寻找像素分类间隔最大的最优分类面,将非线性输入空间的样本映射到高维特征空间进行求解;然后局部核函数选择高斯径向基核函数,全局核函数选择多项式核函数,为了满足训练集中支持向量取值带来的连续性要求,通过组合系数平衡高斯核函数和多项式核函数的权重;接着选择像素的邻域灰度均值作为用于分割的特征,利用不规则度统计图像邻域灰度均值连通区域的离散程度;最后给出了算法流程。实验仿真显示本文算法分割图像清晰,目标区域十分突出,定性分析中指标归一化互相关系数为0.9946,分割时间为0.7512,误割率为0.0237。
To improve the quality of image segmentation, this paper proposed a kernel function of support vector machine algorithm. Firstly, pixel classification intervals were searched through optimal classification, and sample nonlinear space was mapped into high dimensional one; Secondly, Gauss radial basis kernel function was selected as local kernel function, polynomial kernel function was selected as global kernel function , then,combination coefficient was used to balance weight of Gauss kernel function and polynomial kernel function in order to satisfy continuity requirements for training support vector value. Thirdly, neighborhood gray mean pixel was used as features of segmentation, irregularity was used to collect statistics of discrete degree of image gray mean connected region. Finally, the algorithm flow was given. Simulation shows that this algorithm for image segmentation is clear, the target area is very outstanding, the qualitative analysis index normalized cross correlation coefficient is 0.9946, time is 0.7512, and false rate is 0.0237.
出处
《红外技术》
CSCD
北大核心
2015年第3期234-239,共6页
Infrared Technology
基金
河南省社科联项目
编号:SKL-2013-506
郑州市社科联项目
编号:JX20130297
关键词
向量机
核函数
图像分割
vector machine
kernel function
image segmentation
