基于马氏距离和模糊C均值聚类的抠图算法与应用  

Matting algorithm and application based on Mahalanobis distance and the fuzzy C-means clustering algorithm

作　　者：张敏[1] 闵乐泉[1,2] 张群[2] 刘飒[3] 

机构地区：[1]北京科技大学数理学院,北京100083 [2]北京科技大学自动化学院,北京100083 [3]北京农学院计算机科学系,北京102206

出　　处：《北京科技大学学报》2014年第5期688-694,共7页Journal of University of Science and Technology Beijing

基　　金：国家自然科学基金资助项目(61074192);北京市教育委员会科学研究基金资助项目(KM201110020013)

摘　　要：基于马氏距离和模糊C均值聚类算法提出了一种数字彩色图像抠图算法.该算法首先对彩色图像像素的红绿蓝三种彩色分量进行正则化处理;然后在正则化图像背景中选取适当的掩膜作为样本集,计算各像素与样本集之间的马氏距离;再利用模糊C均值聚类算法对计算出的马氏距离进行分类;最后利用填洞操作提高抠图质量.对八幅彩色数字图像进行对比实验,结果显示本算法可以自动抠图,且结果优于马氏距离算法、Grow-Cut算法和正则化线性回归算法的相应抠图效果.ABSTRACT Based on Mahalanobis distance and the fuzzy C-means algorithm, this article introduces a digital color image matting algorithm. First the red, green and blue color components of color image pixels are normalized. Second the appropriate mask as a sample set is selected in the background of the normalized image, and the Mahalanobis distance between each pixel and the sample set is calculated. Third the calculated Mahalanobis distances are classified into two categories using the fuzzy C-means clustering algorithm：the foreground and the background. Finally, the quality of the matting is improved using the filling-hole technique. Eight images have been processed for comparison, the results show that this algorithm can automatically segment these images, and is better than the Mahalanobis distance algorithm, fuzzy C-means clustering algorithm and linear regression algorithm.

关 键 词：图像抠图 马氏距离 模糊C均值聚类 填洞 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]