基于改进分水岭算法的粘连大米图像分割

Image segmentation of adhering rice based on the improved watershed algorithm

大米检测过程中大米图像籽粒粘连给后续大米检测、分析等处理带来较大困难,为此提出了一种有效的改进分水岭分割算法,先对二值图像采用不同的结构元素作极限腐蚀形成不同的距离图,再对这些距离图分别利用分水岭算法得到分水岭图像,最后提取真实分水岭得到最终分割图像,并与其它分水岭算法相比较。实验结果表明,该算法既能更有效地分割粘连颗粒,又能有效地抑制过分割现象。

During rice detection, the touching of rice kernels in image brought a great difficulty on the posterior detection and analysis. Accordingly, an effective and improved watershed algorithm was put forward to take on limit corrosion on the binary image through different structural elements to form different distance maps. And then watershed images were obtained through watershed algorithm. At last, the actual watershed was extracted to ob tain the final segmentation image. As seen from the comparison with the other watershed algorithms, this algorithm could segment more effectively the touched kernels and inhibit the over-segmentation phenomenon.