基于最小类内离散度的改进Otsu分割方法的研究

Improved Otsu thresholding based on minimum inner-cluster variance

提出了一种改进的Otsu阈值分割方法,该算法结合了最小类内离散度与最大类间方差.类内方差越小,类的内聚性就越好,据此提出分类的类内离散测度,综合最大类间方差,定义了新的阈值识别函数.实验结果表明:该方法克服了传统Otsu阈值分割信息不完备的缺陷,具有更强的抗噪能力,分割效果明显.

An improved Otsu thresholding method is proposed which combines the minimum withincluster scatter with the maximum between-cluster variance. Cohesion performance of a cluster increases with decreasing of within-cluster variance. Thus, a new concept was given of scattered measure within clusters and a new threshold recognition function was integrated with the maximum between- cluster variance. Experimental results show that the proposed algorithm with resisting noise overcomes the disadvantage of incomplete information for traditional otsu thresholding segmentation and its segmentation effect is obvious.