基于平均方差和邻域信息的改进型二维Otsu算法

An Improved 2D Otsu Algorithm Based on Average Variance and Neighborhood Information

摘要:

针对传统二维Otsu算法忽略边界信息对图像分割结果有一定影响、计算时间长的问题, 研究提出了一种基于平均方差和邻域信息的改进型算法。新算法先利用中值滤波来重新构建一个二维直方图, 再用平均方差定义一个新的二维阈值选取函数, 最后对图像进行二值化处理。实验结果表明, 与传统二维Otsu法及其快速递推算法相比, 新算法有效提高了分割精度, 减少了算法的运行时间, 其运行时间仅为传统二维Otsu算法的1.55%, 大约是快速递推算法的40.69%。

Abstract:

Aiming at the problems as neglecting edge information and long time computing for traditional 2D Otsu algorithm, an improved algorithm based on the average variance and neighborhood information was proposed. Firstly, 2D histogram was reconstructed with the median filtering, and then a new 2D threshold selection function was defined with the average variance, finally, binarization method was used to segment image. The experimental results show that the improved algorithm can obtain better segmentation results compared with traditional 2D Otsu method and the fast recursive algorithm, the running time is reduced to be 1.55% and 40.69% of that of the traditional 2D Otsu algorithm and the fast recursive algorithm, respectively.