为了解决2维最小误差阈值分割法对呈偏斜分布与重尾分布的图像分割鲁棒性较差的问题,提出一种基于平均中值离差的2维最小误差阈值分割法.考虑到1维直方图呈偏斜分布和重尾分布的图像中,中值是比均值更为鲁棒的灰度级估计量,因而将2维最小误差阈值分割法中的方差用平均中值离差替代;为提高运算速度,将2维算法分解为2个1维算法.结果表明,相比2维Otsu法、2维最小误差阈值分割法等经典算法,基于平均中值离差的2维最小误差阈值分割法对1维直方图呈偏斜分布与重尾分布的图像有更准确的分割效果、更好的鲁棒性.

In order to solve the problem that 2-D minimum error threshold segmentation (METS) method had poor segment robust performance on an image which presents skew distribution and heavy-tailed distribution,an improved 2-D METS method was proposed based on mean absolute deviation from the median.Considering that the median was a more robust estimator of gray level than the mean in 1-D histogram of skew distribution and heavy-tailed distribution,variance in 2-D METS was replaced by mean absolute deviation from the median.In order to improve the computational speed,a 2-D algorithm was decomposed into two 1-D algorithms.Experimental results show that,compared with 2-D Otsu method,2-D METS method and other classical algorithms,the improved 2-D METS method based on mean absolute deviation has more accurate segmentation results and more robust performance for 1-D histogram with skew distribution and heavy-tailed distribution.