结合贝叶斯分类与SLIC的Grabcut彩色图像分割

Grabcut color image segmentation algorithm combined with Bayesian classification and SLIC

针对基于超像素的Grabcut图像分割算法在超像素数目较低时出现分割恶化的现象,提出一种结合贝叶斯分类与SLIC(简单线性迭代聚类)的改进Grabcut分割算法。首先,使用SLIC算法对图像进行聚类,使用聚类后的各像素块的RGB均值作为像素点构建精简的Graph Cuts模型,然后,再使用贝叶斯分类对模型中的像素点进行分类,对进行分类后的像素点进行第二次SLIC聚类,并用各个像素块的均值代替像素点值进行GMM(高斯混合模型)参数估计,最后使用最小割算法得出图模型最优分割。实验结果证明,本算法降低了分割错误,取得了很好的分割效果。

To overcome the phenomenon of deterioration in segmentation which is based on super pixels in the im- age segmentation algorithm of Grabcut when the number of super pixels is low. It put forward a segmentation algorithm which is combined with Bayesian classification and SLIC （Simple Linear Iterative Clustering） to improve Grabcut. First- ly, use the SLIC algorithm for image clustering, after that, use RGB mean value of evey pixel blocks as the pixel dot to form contacted Graph Cuts model. And then use Bayesian classification to classify pixels in the model. Apply SLIC to classify the pixels in the second time. In order to estimate the value of GMM, this algorithm use the mean of the super pixels color value to represent the all pixels color value. Finally it used min-cut algorithm to get the optimal segmenta- tion of graph. The experimental results show that the algorithm reduced segmentation mistake and achieved better image segmentation.