Based on public areas and contours of touching kernels, an approach to search segmentation points was developed, realizing automatic segmentation of touching corn kernels. For two touching kernels, the public area could be obtained in the process of continuous erosion-dilation. Non-closed curve segment was extracted by getting intersection of the public area and one kernel contour. The endpoints of the curve segment were segmentation points. Then the two touching kernels were separated by linking two segmentation points with Bresenham algorithm. For massive touching kernels, every single kernel could be stripped using the similar method. The experimental results on 100 touching kernels images showed that the correction rate of segmentation is 96%. The kernels after segmentation have small deformation and smooth boundaries.