一种基于主动轮廓模型的连接米粒图像分割算法

针对图像中连接米粒分割困难的问题，提出了一种基于主动轮廓模型的分割算法。首先，对籽粒二值图像的欧氏距离变换进行局部极小值检测，并通过形态学膨胀算子合并局部极小值点，在每个籽粒内部只产生一个区域。其次，以这些区域的边界作为初始曲线，在主动轮廓模型的指导下，曲线向籽粒的边界演化，最终将图像中各个米粒分割。试验结果表明，对圆江米、粳米、长江米和黑米4个品种的米粒，基于主动轮廓模型的连接米粒图像分割算法的分割正确率分别达到93.4%、92.4%、88.0%和90.4%，综合准确率为91.05%，比基于分水岭的方法提高了26.7%。因此，基于主动轮廓模型的算法为分割连接米粒图像提供了一种有效途径。

Image segmentation algorithm of touching rice kernels based on active contour model

A new segmentation method based on active contour model was proposed in this paper and was applied to segmenting touching rice kernels in an image. Firstly, the local minimums were detected in the Euclidean distance transform of the binary image of the rice kernels. Those local minimums were merged by the morphologic dilation operator, so that one kernel corresponded to one region. Then, the edges of those regions were taken as initial curves and converged to the true boundary of the rice kernels in the image under the guidance of active contour model. The experimental results showed that the proposed segmentation algorithm based on active contour model got desirable segmentation. Its segmentation accuracy for round glutinous rice, non-glutinous rice, long glutinous rice and black rice reached 93.4%, 92.4%, 88.0% and 90.4%, respectively. That was, its overall accuracy achieved 91.05%, which was 26.7% more than that of the watershed based method. Therefore, the algorithm based on active contour model provides an effective means to separate the touching rice kernels in an image.