改进二维Otsu法和果蝇算法结合的图像分割方法  

Novel method for image segmentation based on improved two-dimensional Otsu and fruit fly algorithm

作　　者：彭启伟[1] 罗旺[1] 冯敏[1] 郝小龙 侯麟 琚小明 

机构地区：[1]南瑞集团公司国网电力科学研究院,南京210000 [2]华东师范大学计算机科学与软件工程学院,上海200000

出　　处：《计算机应用》2017年第A02期193-197,共5页journal of Computer Applications

基　　金：国家电网科技项目(SGSDDK000KJJS1600065)

摘　　要：针对传统二维Otsu方法中分割精度不高和抗噪性能不足的缺陷进行了研究,提出一种改进的二维Otsu图像分割方法。在二维直方图区域叉分的基础上,根据灰度级大小的信息改进联合概率密度,考虑类间方差和类内方差对图像分割效果的作用,保证类间方差越大和类内方差越小,使得目标与背景差别越大和内聚性越高;同时依据目标与背景在图像中所占比例对阈值求取公式进行加权,使得求得的阈值更接近理想阈值,然后采用果蝇优化算法(FOA)搜索最优二维阈值向量。仿真实验结果表明,该方法能得到更准确的分割结果,更好地抑制噪声,同时运行时间更少,达到了快速分割的目的,满足图像处理的实时性要求。As traditional two-dimensional Otsu method is not accurate enough and the filter performance is insufficient,an improved two-dimensional Otsu image segmentation method was proposed. The two-dimensional histogram region was divided by two lines with different angles through the threshold point. The joint probability density was improved according to the gray scale size. Considering the influence of inter-class variance and intra-class variance on image segmentation, the proposed method ensured that the inter-class variance was greater and the intra-class variance was smaller, so that the difference between the target and the background was greater and the cohesion was higher. The threshold was calculated according to target proportion and the background in the image, so that the threshold is closer to the ideal threshold. Then Fruit Fly Optimization Algorithm（ FOA） was used to search the optimal two-dimensional threshold vector. The simulation results show that the proposed method can get more accurate segmentation result, better noise suppression and less running time. The method satisfies the real-time requirement of image processing.

关 键 词：图像分割 二维Otsu阈值 类间方差 类内方差 果蝇优化算法 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]