摘要
对应用广泛的阈值图像分割的最大类间方差法进行了研究,在深入分析了现有的Neighborhood Valley-Emphasis Method存在的问题的基础上,提出一种邻域均值加权最大类间方差的阈值分割法Neighborhood-Mean Valley-Emphasis Method.其核心思想是,对最大类间方差法进行邻域均值加权;根据灰度值出现的次数生成直方图,取每个灰度值的邻域区间,求得灰度值出现次数的邻域均值,同时求得类间方差,用邻域均值加权类间方差,使得阈值在类间方差最大且是直方图的极小点处取得,提高了算法的准确性,另一方面,邻域均值有均值滤波器的效果,增强了算法的抗噪性.实验结果表明,改进的最大类间方差法NeighborhoodMean Valley-Emphasis Method大大降低了因加权溢出而导致算法失效的可能性,扩大了适用范围,增强了鲁棒性.
This paper focuses on widely used Otsu of thresholding segmentation. Considering the problems existed in the current Neighborhood Valley-Emphasis Method, a thresholding segmentation method was proposed in this paper, which uses neighborhood mean to weigh Otsu, i.e. Neighborhood-Mean Valley-Emphasis Method. The main idea is that neighborhood mean weighs Otsu, and obtain histogram according to the occurrence number of gray level, compute the mean of the occurrence number in a neighborhood interval, get between-class variance at the same time, then make neighborhood mean weigh between-class variance, locating the threshold in both maximizing the between-class variance and minimizing the histogram, which improves the accuracy. On the other hand, neighborhood-mean works like mean filter, enhanced antinoise ability. Experiment results showed that modified Otsu, i.e. Neighborhood-Mean Valley-Emphasis Method cut down the possibility of invalidation caused by weighting overflow, inlarged the application sphere, and stengthened robustness.
出处
《小型微型计算机系统》
CSCD
北大核心
2014年第6期1368-1372,共5页
Journal of Chinese Computer Systems
基金
国家自然科学基金项目(11104142)资助
关键词
直方图
最大类间方差
邻域均值谷强调法
多阈值
histogram
Otsu
neighborhood-mean valley-emphasis method
multilevel thresholding
