一种改进的图谱阈值分割算法

Improved Segmentation Algorithm Based on Graph Spectral Threshold

针对图像分割是典型的结构不良问题,将图谱划分理论作为一种新型的模式分析工具应用到图像分割并引起广大学者关注。考虑到现有的图谱阈值法中图权计算方法采用基于欧氏距离的幂指数函数导致其计算量过大的不足,首先采用基于欧氏距离的分式型柯西函数代替基于欧氏距离的幂指数函数提出图权计算的新方法,其次将其应用基于图谱划分测度的图像阈值分割算法中并得到一种改进的图谱阈值分割方法。实验结果表明,该方法的计算量小且对目标和背景相差比例较大的图像能获得满意的结果。

Aiming at the problem of image segmentation with badness structure, the graph cut measure theory is a kind of new type tool of pattern analysis,and it has been applied in the image segmentation field and brings the attention of a lot of scholars. Considering the shortage of graph spectral thresholding method with a great deal of computation because of graph weight computation method adopting power exponential function based on Euclidean distance, the new computation method of graph weights are proposed by mean of replacing power exponential function with fractional Cauchy function based on Euclidean distance,and there are applied in the image thresholding segmentation algorithm based on the measure of graph spectral, The experimental results show that the new method has a small deal of computation and is more suitable to segment the image with the bigger proportition between goal and background,