摘要
针对有限混合模型中参数估计方法对先验假设存在过分依赖和图像数据量大的问题,提出了一种基于抽样的非参数余弦正交序列的图像混合模型分割方法.首先,基于图像的直方图进行分层随机抽样得到样本数据,根据样本数据构建非参数正交多项式混合模型,对于模型的平滑参数采用最小均方差方法进行估计;其次,采用NEM(Nonparametric Expectation Maxi-mum)算法求解混合模型中正交多项式系数和模型的混合比;最后,根据贝叶斯准则进行图像分割.此方法能够克服参数模型的基本假设与实际的物理模型之间存在的差异,实验表明该方法比GMM和Hermite混合模型分割方法分割质量高,而且分割速度快.
Parametric mixture models for image segmentation depend too much on the priori assumptions. To overcome this problem, a non-parametric cosine orthogonal sequence of mixture model for image segmentation method using sampling is proposed in this pa- per. First, sample data is using stratified random sampling Based on the image histogram, the cosine orthogonal sequence base on the image non-parametric mixture models of sample data is designed, the sample data based on the histogram using stratified random sam- pling ,and the mean integrated squared error（MISE） is used to estimate the smoothing parameter for each model; Second, the Non- parametric Expectation Maximum （ NEM } algorithm is used to estimate the orthogonal polynomial coefficients and the model of the weight. This method does not require any prior assumptions on the model, and can effectively overcome the ＂model mismatch＂ prob- lem. The experimental results with the color images show that this method can achieve better segmentation results than the methods of Gaussian Mixture Models and Hermite mixture models.
出处
《小型微型计算机系统》
CSCD
北大核心
2013年第6期1428-1432,共5页
Journal of Chinese Computer Systems
基金
国家自然科学基金项目(60841003)资助
教育部博士点基金项目(20113227110010)资助
吉林教育厅"十二五"科学技术研究项目(吉教科合字[2013]第448号)资助
江苏省博士创新基金项目(CX10B_274Z)资助
关键词
非参数混合模型
图像分割
平滑参数
正交多项式
non-parametric mixture models
image segmentation
smoothing parameter
orthogonal polynomial
