摘要
为克服经典区域增长算法中生长规则以及特征选取的困难,提出了基于高斯混合模型的多区域并行区域增长图像分割算法。首先交互选择多个不同区域的种子点,并利用交互式选择的属于每个区域的子块得到混合模型的个数;然后利用最大期望估计混合模型参数作为区域增长的初始参数,并在增长过程中不停地调节模型参数。为了避免初始种子点位置选择对算法性能的影响,采用了多区域并行竞争增长策略。仿真实验获得了较好的分割效果,表明所提出的算法是合理可行的。
To overcome the difficuhy of thrcshold selection and region growing rule existing in conventional region growing image segmentation algorithm, a multiple regions parallel growing algorithm based on Gaussian mixture model (GMM) is proposed. Initial .seeds and the blocks belong to the object region in each object are selected by interactive manual operation, so the number model and initial parameters of GMM can be easily obtained. Then the parameters of global GMM are estimated by using expectation maximization algorithms. In region growing processing step, the GMM is updated according to the results of each growing region. In order to make results independent of processing order and the initial growing seeds, multiple region parallel growing is employed. Experimental results show that this algorithm is feasible, and it performs better than conventional region growing algorithm.
出处
《光学技术》
EI
CAS
CSCD
北大核心
2006年第6期814-816,共3页
Optical Technique
关键词
区域增长
高斯混合模型
图像分割
region growing
Gaussian-mixture model (GMM)
image segmentation