Abstract: In the process of region merging, manually setting the weight coefficient of the color similarity and the boundary distance of adjacent regions affects the accuracy and automation of segmentation method. For this issue, a novel segmentation algorithm based on multi-level region merging is proposed, which can set the weight coefficient according to the boundary feature of adjacent regions. And so the color similarity and the border distance are used adaptively. An image is over segmented by mean shift algorithm. The initial regions are merged to form the segmentation result by a new region merging strategy. Experimental results on color images verify that the method outperforms the referenced ones.

Key words: color image segmentation, mean shift algorithm, adaptive, region merging

摘要： 在区域合并过程中，手工设置颜色相似性和边界距离的权重极大地影响了分割的精度和自动化。针对这一问题，提出了一种新的基于区域分级合并的彩色图像分割算法。该方法能够根据邻接区域的边界特点设置权重因子，从而自适应地融合区域的颜色相似性和边界距离。使用均值漂移算法对图像进行初始分割，将原图像分割为具有较好边界的同质区域；通过计算区域相似度对区域进行分级合并。多幅彩色图像的分割实验结果证明，所提算法优于传统的基于区域合并的方法。

关键词: 彩色图像分割, 均值漂移算法, 自适应, 区域合并