快速递归多阈值分割算法

Fast recursive multi-thresholding algorithm

针对强调波谷邻域算法在目标区域相对于背景区域较小且其之间的波谷特征并不十分明显的情况下,无法获得正确阈值的问题,提出了一种基于波谷邻域信息和波谷波峰相对特征的全局阈值分割算法。本算法在最大类间方差(OTSU)算法的基础上以直方图中波谷邻域灰度值和波谷波峰灰度值的相对关系为权值,改善最大类间方差算法定位阈值的准确性,使算法所确定的阈值在直方图中具有较小的波谷波峰比值,即使最优阈值定位到与临近波峰具有较大高度差的波谷灰度值。为提高分割效率,本文以前述算法为基础,采用递归单阈值方式进行图像的多阈值分割。实验证明,对强调波谷邻域算法存在的问题本算法有明显的改善,且在多阈值分割的效果及运行时间方面本文算法均具有十分良好的表现。

The Neighborhood Valley-emphasis method can not get the right threshold value in some cases,such as the valley feature between the target and background is not very distinct.In order to solve this problem,aglobal thresholding method is proposed.This method is based on the gray information around the valley-point neighborhood and the relative characteristics between the valley point and its adjacent crest-point.The proposed method weights the objective function with the gray information around the valley-point neighborhood and the relation between the valley-point and its adjacent crest-point.It improves the accuracy of the threshold obtained by OTSU.The optimal threshold got by the proposed method has less valley-to-crest ratio.In other word,the valley gray is taken as the optimal threshold,which has larger height difference with it adjacent crest-point.Inorder to improve the efficiency,a recursive single threshold method based on the aforesaid algorithm is used to achieve the image multi-threshold segmentation.Experiment results show that the proposed method has great segmentation performance and low time complexity.