基于Parzen窗技术和信息熵的图像分割阈值选取新方法PWET

The Novel Image Thresholding Method PWET Based on Parzen Window Technique and Information Entropy

传统的阈值选取方法是以图像的灰度直方图做为像素灰度值的概率密度函数的近似估计,然后利用此密度函数来构造目标函数,通过在灰度值中搜索使目标函数最大化(最小化)的灰度值做为最优全局阈值,从而实现图像的分割.为了克服以直方图(灰度值出现的频率)替代灰度值的概率分布不够准确的缺陷,提出了一种基于parzen窗技术和信息熵的阈值选取新方法PWET.该方法以图像的像素坐标集为样本空间,利用parzen窗技术估计图像灰度值的空间概率分布,再结合信息熵来构造新的目标函数,通过在灰度值范围内搜索使目标函数最大的灰度值作为最优全局阈值.通过将PWET方法和传统的KSW熵方法进行比较实验,结果表明PWET方法对图像分割更有效.

The traditional thresholding method first used histogram as the approximate estimation of probability density function of the pixel gray level values,then used this density function to construct the object function,and at last searched the optimal global threshold in the scope of gray level to maximum （minimum） the object function to achieve the image segmentation. In order to overcome the defect of the inaccuracy that substituted the histogram （the frequency of gray level values） for the probability distribution of gray level values,a novel image thresholding method PWET which is based on parzen-window technology and information entropy is proposed. The method first acted the image pixel coordinates as the sample space,used parzen-window algorithm to estimate the spatial probability distribution of image gray level values,then combined with the information entropy to construct a new object function,and at last searched the optimal global threshold in the scope of gray level to maximum the object function. Through the experiment which compared the method of PWET with the traditional method of KSW entropy,it shows that the PWET method has a better performance of image segmentation.