喷墨印花纹理图像的期望最大化聚类分割算法  

Clustering segmentation algorithm based on expectation maximization for ink-jet printing image

作　　者：周佳男[1] 冯志林[2] 朱向军[2] 

机构地区：[1]浙江商业职业技术学院信息技术学院,浙江杭州310053 [2]浙江工业大学之江学院,浙江杭州310024

出　　处：《机电工程》2013年第8期1029-1032,共4页Journal of Mechanical & Electrical Engineering

基　　金：国家自然科学基金资助项目(61070113);浙江省自然科学基金资助项目(LY13F020027;LY13F010010);浙江省教育厅科研资助项目(Y201225381)

摘　　要：针对聚类分割算法对喷墨印花纹理图像存在的局限性,提出了一种结合期望最大化(EM)的喷墨印花纹理图像聚类分割算法(CSA)。首先,将空间相关性引入聚类分割中,利用自回归模型表征纹理同质区域;然后,为了提高分割模型参数估计的精度,将分块标定机制引入期望最大化算法中,实现了参数极大似然估计的迭代算法,解决了不完全数据参数估计问题;最后,利用数据集分块并进行聚类,使同类元素具有较高的相似度,从而对图像中的像素进行了归类划分,并将得到的结果进行了合并,实现了目标图像的正确分割。实验结果表明,和传统的聚类分割算法相比,该算法能更好地解决喷墨印花纹理图像的分割问题。Aiming at the limitation for ink-jet printing images of clustering segmentation algorithm, a new image segmentation algorithm based on clustering technology combined with expectation-maximization was proposed. Firstly, the proposed segmentation algorithm was featured by incorporating spatial correlations, and the autoregressive model was applied to generate primitive homogeneous texture regions. Secondly, in order to improve the precision of parameter estimation for the segmentation model, a block-labeling mechanism was further introduced into the expectation-maximization algorithm,which can solve the maximum likelihood parameter estimation and deal with the parameter estimation of incomplete data. Finally, the image data sets were divided into some pieces of data block and went through clustering to make similar elements have a high similarity. After block clustering, pixels in the texture image were classified division for the following task of incorporating received result to achieve the correct segmentation of the target texture image. Experimental results show that, the proposed algorithm can provide a significant improvement over other common clustering segmentation methods to solve the ink-jet printing image segmentation problem.

关 键 词：喷墨印花纹理 期望最大化 聚类分割算法 

分 类 号：TP317.4[自动化与计算机技术—计算机软件与理论]