基于最小偏态指标的图像阈值分割新技术

Image Segmentation New Technique Based on Minimum Bias-normal Distribution Index

本文在假设图像目标和背景像素灰度值均服从正态分布的前提下 ,提出基于最小偏态指标的图像分割新技术 ,该方法引入数理统计中的偏态指标作为图像分割的准则函数 ,利用图像直方图计算各灰度级下的偏态指标值 ,其最小值即对应于最佳分割阈值 .本文对最小偏态指标法进行了详尽的图像分割验证 ,并与 Otsu法、最大熵法和最小误差准则法进行了详细比较 ,结果表明本文方法具有分割精度高、计算速度快以及对目标大小影响小等优点 。

In this thesis, in case that the object and the background in image obey all normal distributions, the image segmentation new method based on minimum bias-normal distribution index is put forward. In new method, the bias-normal distribution index in mathematical statistics is introduced, and is regarded as image segmentation criterion function. The image histogram is used to compute the bias-normal distribution indexes of all gray grades, and the optimal threshold is determined at the minimum value of bias-normal distribution indexes. In the paper, the minimum bias-normal distribution index method is testified, and is compared with the Otsu method, the maximum entropy method, and the minimum error criterion function method in detail. The results show that the minimum bias-normal distribution index method which is put forward in this thesis has these advantages such as high segmentation precision, fast computation speed, and minor influence by the object's size.