基于直方图估计和形状分析的沥青路面裂缝识别算法  

Asphalt pavement crack recognition algorithm based on histogram estimation and shape analysis

作　　者：徐志刚[1] 赵祥模[1] 宋焕生[1] 雷涛[1] 韦娜[1] 

机构地区：[1]长安大学信息工程学院,西安710064

出　　处：《仪器仪表学报》2010年第10期2260-2266,共7页Chinese Journal of Scientific Instrument

基　　金：国家自然科学基金(60902075);教育部"长江学者和创新团队发展计划"创新团队基金(IRT0951);教育部新世纪优秀人才支持计划(NCET-05-0849);长安大学中央高校基本科研业务费(CHD2009JC114;CHD2010ZY007)资助项目

摘　　要：提出了一种基于直方图估计和形状分析的沥青路面裂缝识别算法,该算法首先将1024×1024像素大小的路面裂缝图像分为256个64×64像素大小的子块,然后采用直方图估计的方法获得每个子块图像原始直方图的混合高斯拟合函数,两个高斯函数的交叉点即是每个子块图像的最优分割阈值。利用该阈值对整幅图像进行二值化后,在两种尺度条件下采用形状分析方法对子块二值图像进行快速分类和"野点"删除,最终实现了裂缝区域的精确定位。试验结果表明:本文提出的阈值分割方法应用于裂缝图像分割,其性能要优于极小误差法、Ostu阈值法、最大熵法等经典算法;采用形状分析对分割后二值化图像进行后续处理,可实现裂缝区域的快速、准确定位。An asphalt pavement crack recognition algorithm based on histogram estimation and shape analysis is proposed. The original pavement image （1 024×1 024 pixels size） is first divided into 256 cells （64×64 pixels size）. With histogram estimation method, a mixed-Gaussian fitting function is obtained for the original histogram of each cell; and the intersection of two Gaussian functions is taken as the optimal segmentation threshold. After image binarization using the threshold value, these binary cells are classified into two different types and the ＂wild spots＂ are deleted using shape analysis theory under two scales. So the cracks are located precisely on the whole image. Experiment results show that the proposed thresholding method is superior to those classical algorithms, such as minimal error method, Ostu threshold segmentation method and maximum entropy method in crack image segmentation; and with shape analysis, the crack region can be located quickly and accurately.

关 键 词：图像处理 路面裂缝检测 阈值分割 直方图估计 形状分析 神经网络 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]