基于组件树滤波及快速区域合并的分水岭分割算法

摘要

An image segmentation algorithm combined with image filtering by using component tree and fast region merging was proposed to deal with over-segmentation. Firstly, the component tree was used to represent the gradient image, the relative potential energy and its properties were computed according to the ordered extremum. Then the gradient image was filtered to reduce the local minima in pre-processing stage. Then, the watershed algorithm was applied to the filtered gradient image to get the pre-segmentation result Finally,a fast region merging algorithm was used to merge the pre-segmentation regions based on the perfect scene criterion to get the final segmentation. Experimental results show that image filtering by using component tree can reduce the local minima due to the noise and reduce the o-ver-segmentation. It can help improve the region merging accuracy and processing speed greatly.%针对分水岭算法存在过分割的问题,提出一种结合组件树滤波及快速区域合并的图像分割算法.该算法在图像预处理阶段利用组件树来表示梯度图像且根据顺序极值计算分水岭的相对势能和属性,并对其进行滤波,从而减少梯度图像中的局部极小值.对滤波后的梯度图像进行分水岭初始分割,然后利用完美场景准则对初始分割结果进行快速区域合并.实验结果表明,采用组件树对梯度图像进行滤波能够减少由于噪声而产生的局部极小值,大大减少了分水岭初始分割区域数量,提高了区域合并的准确性,加快了合并速度.