基于GrabCut的磨粒图像分割方法研究

Segmentation of Wear Particle Image Based on GrabCut

提出基于GrabCut算法的彩色磨粒图像分割方法。在CIELab颜色空间的L、a、b通道采用大津阈值法(OSTU)进行分割,对分割结果进行"L与(a或b)"处理,并对处理结果进行形态学腐蚀和膨胀,得到标记出磨粒区域、可能的磨粒区域和背景区域的标记图像。根据标记图像分别对磨粒区域和背景区域的高斯混合模型(GMM)参数进行初始化,构建带权值的无向图映射原图像并建立包含区域项和边界项的Gibbs能量方程。通过迭代运算调整各像素点所属类别和高斯混合模型的参数,采用mincut/maxflow算法极小化能量函数,当能量趋于收敛时即可得到最终的分割结果。

In this paper,a wear particle image segmentation approach based on GrabCut is proposed.OSTU is used to do the image segmentation in the L,a and b channels of the CIELab color space.Segmentation results are processing by “L and ( a or b )”,and the processing results are morphologically corroded and expanded to obtain the marked image.The wear particle region,the possible wear particle region and the background region are marked by the marked image.The parameters of the Gaussian mixture model (GMM) of the wear particle region and the background region are initialized according to the marked image,which is used to map the image with weighted undirected graphs and establish a Gibbs energy equation containing region terms and boundary terms.Then,the iterations are used to adjust the parameters of each pixel and the parameters of the mixed Gaussian model and min/maxflow algorithm is used to minimize the energy function,and the segmentation result can be obtained when the energy tends to converge.