摘要
提出一种阈值自适应、EM方法估计GMM参量的图像分割算法,能够根据图像的内容结合区域和边界两方面的信息自适应地选择阈值,精确地进行图像边界分割.算法首先提取图像的边界,然后根据边界的直方图计算图像的可分割性,由可分割性确定EM方法的阈值进行GMM分割,最后合并图像的近似区域.实验数据表明,相比其它图像分割算法,以及固定阈值的传统EM算法,本算法的分割结果更为准确.
A new segment algorithm which was self adaptive to the image content, combining patch-based information with edge cues under a probabilistic framework was presented. Edges were detected firstly. Then a histogram and the segmentable measure of the imgae were computed. I.ater EM algorithm was adopted to estimate the mixture of multiple Gaussians which was built as a statistical model on spatial features. Lastly the adjacent regions with similar properties are united to one. The novelty of this algorithm is that the threshold was computed by segmentable measure which was adaptive to image. Some experimental results are qualitatively and quantitatively evaluated on a large data-set of natural images by rand index （RI） and ground-truth,and it shows that the proposed method has an outstanding effect.
出处
《光子学报》
EI
CAS
CSCD
北大核心
2009年第6期1581-1585,共5页
Acta Photonica Sinica
基金
国家重点学科(G708)
上海市重点学科(B67)资助
关键词
图像分割
混合高斯模型
期望最大算法
自适应阈值
Image Segment
Gaussian mixture model （GMM）
Expectation Maximization （EM） algorithm
Self-adautive threshold
