一种改进的医学图像分水岭分割算法

Improved watershed segmentation algorithm for medical image

为了提高医学图像分割的准确性,针对分水岭分割算法中的过分割问题,提出了一种改进的医学图像分水岭分割算法。该算法首先在分水岭变换前进行预处理初步分割,主要包括多尺度形态学滤波、多尺度梯度算子计算、自适应标记提取以及分水岭变换;然后在初步分割变换后,通过基于邻接图的区域灰度相似性与边界相似性相结合的合并准则,对分割后的区域进一步合并。实验结果表明,新算法有效地解决了分水岭算法的过分割问题,具有较强的抗噪性能和边缘定位能力,能够满足医学图像的分割要求。

In order to improve the accuracy of medical image segmentation, aiming to the overdivided problem of the watershed segmentation algorithm, this paper proposed an improved watershed segmentation algorithm for medical image. Firstly, it pre- treated the initial segmentation before the watershed transform, mainly including multi-scale morphological filtering multi-scale gradient operator, adaptive marker extraction and watershed transform; and then after the initial segmentation, further merged the segmented region according to the merger criterion of the combination of the adjacency graph regional gray similarity and boundary similarity. The experimental results show that, the new algorithm effectively solves the overdivided problem of the watershed algorithm and has strong robust to noise performance and the edge location capability, can meet the requirements for medical image segmentation.