摘要： 针对分水岭算法预处理阶段的标记提取问题，提出一种基于人类视觉特性的梯度骨架标记提取算法。通过标记梯度图像的谷点组成贯通整幅梯度图像的骨架，在不同彩色空间中根据背景亮度自动选取阈值，并从骨架中剔除大于该阈值的鞍点，使骨架离散为各自连通的小骨架群，将每个小骨架作为独立标记。实验结果表明，该算法能够解决传统算法参数选取困难的问题，在抑制分水岭算法过分割边缘的基础上获得更好的分割结果。

关键词: 分水岭算法, 标记提取, 梯度骨架, 人类视觉特性, 韦伯曲线

Abstract: To resolve the problem of marker extraction during the preprocessing of watershed algorithm, a gradient skeleton marker extraction method based on human visual characteristics is proposed in this paper. It marks all the valley point of gradient image to build up a connected skeleton. It chooses automatic threshold according to difference background in difference color space and picks out and abandons the saddle point whose gray value is larger than the saddle point threshold. After that, the skeleton is separated to some little discrete skeletons, each of them is a marker. Experimental results indicate that this algorithm can avoid the difficult of parameter initialization in traditional marker extraction algorithm and well suppress the over-segmentation to get a more satisfactory segmentation result.

Key words: watershed algorithm, marker extraction, gradient skeleton, human visual characteristics, Weber curve

中图分类号: 

• TP391.41