基于快速递推模糊2-划分熵图割的红外图像分割

考虑现有图割算法没有充分考虑红外图像的模糊特性,分割精度和运行效率低的缺点,提出了基于快速递推模糊2-划分熵图割的红外图像分割算法以实现复杂背景下红外图像的自动高效分割。该方法利用图像感兴趣区域的最大模糊熵信息设计图割能量函数的似然能,基于局部最大模糊2-划分熵值迭代检测出包含图像最大信息的感兴趣区域来确保提取目标信息的完整性。为了提高最大模糊熵寻优的效率,引入时间复杂度为O(n2)的递推算法,将模糊熵计算转化为递推过程,并保存所有递推的熵函数值用于后续的穷举寻优。针对确定的感兴趣区域,利用该区域最大模糊2-划分时隶属度函数分布设置图割能量函数的似然能,从而充分考虑图像的模糊特性。对分割结果与几种常用的算法进行了视觉比较及运行时间,错分率,F指标的量化分析。结果表明:该算法分割精度F值高达95%,运行时间较其他常用算法至少缩短了72%,基本满足自动红外图像分割对精度、效率和鲁棒性的要求。

Infrared image segmentation based on graph cut of fast recursive fuzzy 2-partition entropy

Most of existing Graph Cut(GC) algorithms have not considered the fuzzy feature, poorer segmentation precision and lower operating efficiency of infrared images sufficiently. So this paper proposes an infrared image segmentation method based on the GC of fast recursive fuzzy 2-partition entropy to implement the automatic segmentation of an infrared image in complex backgrounds. The information of the maximum fuzzy entropy from a Region of Interest(ROI) was used to set the likelihood energy of the GC. The ROI containing the maximum image information was detected by the iterative condition scheme based on the local fuzzy entropy values to ensure the completeness of the extracted target information. To improve the searching efficiency of the maximum fuzzy entropy, a recursive algorithm with time complexity O(n²) was presented to convert the computation of fuzzy entropy into a recursive process, and all the values of recursive entropy function were cached for the succeeding exhaustive optimization. For certain ROI, the likelihood energy of the GC energy function was set by the maximum fuzzy 2-partition membership functions of the ROI. By this way, the fuzzy feature of the infrared image can be considered sufficiently. The experimental analysis of the proposed algorithm on visual results, running time, misclassification error as well as F values were compared to those of several common algorithms. A plenty of experimental results indicate that the segmentation precision of proposed algorithm is up to 95% and the running time is 72% shorter than those of compared algorithms. It satisfies the requirements of automatic infrared image segmentation for higher precision, rapid speed, as well as stronger robustness.