二阶广义概率的二维Otsu阈值分割

针对传统二维Otsu阈值法分割结果不够准确和普适性较差等问题,提出了一种基于二阶广义概率的二维Otsu快速图像分割方法。首先用改进的邻域模板构建二维直方图并将其对应的二元概率分布进行修改得到二阶广义概率以便提高分割性能;然后对二维直方图主对角线上的目标和背景两区域的Otsu公式中对应量准确取值,使阈值选取更准确,并通过调整广义概率的参数来提高算法的通用性;最后对二维直方图进行分析得到Otsu计算特性,利用此特性导出新型、快速的递推算法来降低计算复杂度。实验结果表明:与当前二维Otsu法相比,本文提出的方法不仅分割更准确和通用性更强,而且占用的存储空间和运行时间都更少。

Two-dimensional Otsu image thresholding based on second order generalized probability

Aiming at the problems of inaccurate segmentation and bad universality in traditional two-dimensional(2-D) Otsu thresholding methods,a fast 2-D Otsu image thresholding method based on the second order generalized probability(SOGP) was proposed.First,a 2-D histogram was created with the improved neighborhood and the two-variable probability distribution of it was modified to the second order generalized probability to obtain better segmentation performance.Then,the values of objects area and background area in the 2-D histogram main-diagonal district in the Otsu method were precisely calculated to get a more accurate threshold,and the parameter of SOGP was selected to improve the method’s universality.Finally,a 2-D histogram was analyzed to get Otsu computing features,and a new recursive algorithm was inferred with the features to reduce the computational complexity.Experimental results showed that the proposed method could not only achieve more accurate segmentation results and more applicability,but also required much less memory space and running time,compared to the current 2-D Otsu thresholding methods.