面向大数据图像处理的尺度空间挖掘算法及应用  

Big Data Image Scale Space Mining Algorithm and Its Application

作　　者：刘利钊[1] 洪江水[1] 刘莉莉[1] 朱顺痣[1] 尹华一[1] 许华荣[1] 

机构地区：[1]厦门理工学院计算机与信息工程学院,福建厦门361024

出　　处：《上海交通大学学报》2015年第11期1731-1735,共5页Journal of Shanghai Jiaotong University

基　　金：国家自然科学基金项目(61273290;61373147;61503316);中国博士后科学基金面上项目(2013M530261);厦门市科技计划高校创新项目(2014S0048);厦门市科技计划高校创新项目(3502Z20131158);福建省科技厅重大专项(2013HZ0004-1);横向科研课题(Hj13011;Hx13032;C2014060;D2014063;0900300182)

摘　　要：针对大数据挖掘及融合分析的基础理论问题,提出具体可操作的大数据尺度空间挖掘及融合分析方法.根据无限小的质点型热源在物理空间中的热量扩散规律,推导出大数据非线性尺度空间扩散方程.以各向异性的方向可调小波为热源函数,构建了各向异性的大数据非线性尺度空间;以各向同性高斯核函数为热源构建了各向同性大数据非线性尺度空间,以该空间为核心设计了各向异性特征挖掘算法.实验结果显示以各向异性小波热源空间为核心的挖掘算法可以进行分级差异化检测、特征强化、特征整合并最终锁定稳定特征,以各向同性高斯大数据空间为核心的检测算法可以对数据进行大量泛化特征检测.以光强色调饱和度、主成分分析、加权函数为热源构建了单热源和组合式热源大数据非线性空间,进而设计了融合算法并进行了多种类型的挖掘和融合实验;实验结果显示在清晰度、相关系数、信息熵、标准差、特征点数、稳定特征点数等指标方面,组合式大数据非线性尺度空间优于单热源融合空间及传统方法.综合上述可以得出:大数据非线性尺度空间具有强扩展性、开放性和适应性.The paper focused on the basic theory of big data mining and fusion,according to the heat diffusion law of the infinitely small particles heat source in the physical space deduced the big data nonlinear scale space diffusion equation.Different types of nonlinear big data scale space of heat source could be obtained.The anisotropic wavelet function was made as a heat source to build an anisotropic big data nonlinear scale space,the isotropic Gaussian kernel was constructed as a heat source to build isotropic big data nonlinear scale space,the feature mining algorithm was designed,and experiments were conducted.The experimental results show that the heat source anisotropic wavelet space can give differentiated detection,feature enhancement,feature integration and eventually locking stability characteristics.The isotropic Gaussian kernel big data space heat source detection algorithm can detect lots of generalization features.Using the intensity hue saturation,principal component analysis and the weighting function as single heat source and hybrid heat source,a nonlinear big data scale space was built,a fusion algorithm was designedand various image fusion experiments were conducted.The results indicate that the proposed algorithm is superior in the clarity of the correlation coefficient,information entropy,standard deviation,feature points,feature points and stability indicators,hybrid big data space to single big data space and the traditional methods.

关 键 词：大数据非线性尺度空间 热扩散 扩散方程 特征挖掘 图像融合 

分 类 号：TH311[机械工程—机械制造及自动化]