In view of the inaccurate segmentation in the traditional two-dimensional (2-D) Otsu' s thresholding method mainly owing to the supposition that the sum of probabilities of main-diagonal districts in 2-D histogram is approximately one, afast and precise 2-D Otsu's image thresholding method is presented in this paper. A 2-D histogram was created with the select neighborhood and Otsu's method was used on the 2-D histogram in order to obtain better segmentation performance. The probabilities and the mean gray levels in the objects and the background of 2-D histogram main-diagonal districts were calculated preciselyto get a more accurate threshold.The 2-D histogram projection was analyzed to get its features, three theorems were proved,and a novel recursive approach was inferred with the features and the theorems to reduce the computational complexity. Experimental results show that the proposed method achieves more accurate segmentation results and more robust anti-noise than the current 2-D Otsu' s thresholding methods, and that its running time is much less, almost the same as that of the current Otsu's recursive algorithm based on 2-D histogram oblique segmentation.%传统二维Otsu法主要由于对二维直方图采用主对角线区域概率和近似为1的假设等原因,以致分割结果不够准确.针对此问题,提出了一种二维直方图准分的Otsu快速图像分割方法.(1)准确选择邻域模板构建二维直方图并将Otsu阈值法用于此直方图上以便提高分割性能；(2)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；(3)对二维直方图投影进行分析得到其特性,并证明三个定理的存在,利用此特性和三个定理导出新型、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阈值法相比,本文提出的方法分割更准确和抗噪性更强,而且其运行时间少,与当前二维Otsu斜分递推算法的运行时间相近.
展开▼
