一种用于图像分割的等周算法改进  

Improved isoperimetric algorithm for image segmentation

作　　者：汪云飞[1] 毕笃彦[1] 黄飞[1] 

机构地区：[1]空军工程大学工程学院,陕西西安710038

出　　处：《西安电子科技大学学报》2012年第2期87-94,212,共9页Journal of Xidian University

基　　金：国家部委重点实验室基金资助项目(9140C610301080C6106);航空科学基金资助项目(20101996009)

摘　　要：为了解决等周分割算法用于图像处理时抗噪性较弱,迭代效率低的不足,提出了一种新的改进思路,将图中任意结点的密度值类比为矩形的长和宽,在矩形周长一定的条件下,用矩形当前面积与最大面积的比值设计出一种新的权重函数,直接避免了归一化处理的问题;同时充分利用次优等周率参与迭代分割,提高了算法的迭代效率.分析和实验结果表明,在保持原算法分割质量的前提下,所提算法在有效提高原算法抗噪性的同时,进一步减少了迭代运算量的10%～30%,提升了算法的性能.An in-depth analysis is conducted on the graph-theory-based-isoperimetric-partition-algorithm（GTBIPA）,which has such drawbacks as low noise immunity and low iterative efficiency when applied to image processing,and a new approach is presented.In this new approach,the intensity of arbitrary nodes in graph are analogized as the length and width of a rectangle.On the condition of a constant perimeter of a rectangle,a new weighting function is designed by using the ratio between the present and the maximum area of the rectangle,so that the normalized problem is directly avoided;and by the participation of suboptimal isoperimetric ratio in iterative partitioning,the iteration efficiency of the algorithm is improved.Experimental results show that this new algorithm can efficiently enhance the performance of the GTBIPA by improving its noise immunity and reducing the number of iterations by 10%～30%.

关 键 词：图论 图像分割 等周算法 权重函数 次优等周率 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]