基于梯度修正和区域合并的分水岭分割算法

Watershed segmentation based on gradient modification and region merging

分水岭是一种有效的图像分割方法,但存在过分割现象,为此提出了一种改进的分水岭分割方法,该方法首先利用形态学算子得到梯度图像,然后利用形态学混合开闭重构算子进行梯度修正,去除了易造成过分割的区域细节和噪声,接着采用基于标记的分水岭算法进行分割,有效地克服了过分割现象。为了得到更好的分割效果,提出了基于区域一致性和边界曲率光滑性相结合的区域合并准则,对分割后的图像进行有效的合并,该方法能够很好的解决过分割现象,并且产生更有意义的分割效果。通过多组实验,并且和传统的分水岭算法进行对比,得到了满意的效果,结果表明了该方法的有效性。

In order to reduce the over segmentation of the traditional watershed algorithm, an improved marker based watershed image segmentation method. First, morphological gradient image is obtained. Second, morphological techniques called opening-by recon- struction and closing-by-reconstruction are used to reconstruct gradient image. Finally, the watershed algorithm is applied to the modified gradients by the markers to reduce effectively the over segmentation. Furthermore, incorporate boundary curvature ratio, region homogeneity and boundary smoothness into a single new merging criterion is proposed to improve the over-segmentation of marker-controlled watershed segmentation algorithm. Experimental results demonstratethe merits of this method.