Abstract: In view of the obvious shortage of several segmentation methods, such as the gray-the average gray level 2D histogram method, the average gray level-gradient method and 2D histogram oblique segmentation, which are low region homogeneity, low region contrast, and cannot segment accurately enough under the influence of high intensity Gauss noise. This paper proposes a method of maximum between-cluster variance correlation of average gray level-local variance 2D histogram, which uses local variance that not only takes the discrete degree of each pixel point and the center of pixel points into consideration, but also decreases the influence affected by noise. This paper uses a fast recursion algorithm to reduce the amount of calculation. The experimental results show that the method is better than the gray-the average gray level method and the average gray level-gradient method, and has more accurate segmentation results, higher region contrast, region homogeneity, and better anti-noise property.

Key words: image segmentation, high intensity Gauss noise, maximum between-cluster variance, 2D histogram, average gray level-local variance

摘要： 为了提高图像在受到高强度高斯噪声影响下的分割效果，针对传统的二维直方图灰度-平均灰度法，平均灰度-梯度法，二维Otsu斜分法等方法一致性低、对比度低和分割不够准确的情况，现提出一种改进的二维直方图灰度-局部方差的方法。局部方差不仅综合考虑了各像素点与中心像素点数据的离散程度，而且降低了图像受噪声干扰的影响。为了提高分割速度，减少计算量，使用了快速递推算法。实验结果表明该方法比传统的Otsu灰度-平均灰度法和平均灰度-梯度法具有更好的分割效果、一致性和对比度更高。

关键词: 图像分割, 高强度高斯噪声, 最大类间方差法, 二维直方图, 平均灰度-局部方差