三维Otsu阈值分割方法的递推算法

一维和二维Otsu法是常用的阈值分割方法,二维Otsu法对含噪图像的分割效果要优于一维Otsu法,但它存在的问题是"对象区域和背景区域上的概率和近似为1"的假设的普适性不够,鉴于此景晓军等人提出了三维Otsu法及其递归算法,使得对低对比度、低信噪比的图像有较好的分割效果.本文指出了景晓军等人给出的递推公式的错误,并进行了修正.鉴于他们给出的递推公式并不是真正意义的递推公式,本文给出了新的递推公式.实验结果表明,本文提出的递推公式进一步减少了运算时间.另外,本文还指出三维Otsu法对于叠加了混合噪声的图像有较好的分割效果.

Recursive Algorithm for Three-Dimensional Otsu's Thresholding Segmentation Method

One-dimensional(1-D)and two-dimensional(2-D)Otsu's methods are two common thresholding segmentation methods.2-D Otsu's method is superior to 1-D Otsu's method for images with noise.A basic assumption for 2-D Otsu's method is that "the probability sum of the object and background regions is close to 1",this assumption could not be effective in some cases,thus Jing et al.introduced three-dimensional(3-D) Otsu's method and gave a recursive algorithm.3-D Otsu's method behaves well for images with low SNR and low contrast.In this paper,the mistakes in Jing's recursive formulas are pointed out and a corrected version is given.Because the existed recursive algorithm can not be considered as a real recursive algorithm,we propose a new recursive algorithm.Experimental results show that the computing time of our method is less than that of Jing's method,in addition we point out that 3-D Otsu's method can obtain better segmentation results for images with mixture noises.