融合多结构滤波和多尺度重建的分水岭算法

摘要

In order to solve the problem of over-segmentation existing in traditional watershed algorithm, a new method combining multi-structure morphological opening-closing filter and multi-scale morphological opening-closing recon-struction is proposed. It applies 12 different structure elements to original image and uses multi-scale technology for gradi-ent reconstruction, integrates different reconstructed gradient images into result gradient. Extended-minima transform (H-minima)is used to extract markers from result gradient. It modifies original gradient with the markers so that an improved gradient is produced, and then the watershed algorithm is performed. The experimental results show that this method not only can effectively suppress the over-segmentation of watershed algorithm, but also can prevent the occurrence of under-segmentation by adjusting the parameters in the process of segmentation, which can achieve ideal results for differ-ent requirements.%为了解决传统分水岭算法中存在的过分割问题，提出一种结合多结构形态学开闭滤波和多尺度形态学开闭重建的新方法。该方法用12种不同结构的结构元对原始图像进行开闭滤波，接着对梯度图像应用多尺度开闭重建，通过加权融合不同的重建梯度为最终的梯度。在重建后的梯度上应用扩展最小变换技术（H-minima）以提取标记，用所得的标记修正原始梯度，在修改后的梯度上进行分水岭分割。实验结果表明：该方法不仅能有效抑制分水岭算法中的过分割，而且通过调节分割过程中的参数，还能防止欠分割的发生，对于不同的需求均可得到理想的分割效果。