基于多特征的路面裂缝目标提取方法  

Pavement Crack Object Extraction Method Based on Multi- features

作　　者：姜吉荣[1] 陈小惠[1] 

机构地区：[1]南京邮电大学自动化学院,江苏南京210023

出　　处：《计算机技术与发展》2016年第5期173-178,共6页Computer Technology and Development

基　　金：江苏省科技支撑计划项目(BE2011843);南京邮电大学实验室建设项目(2012XZZ06)

摘　　要：根据路面裂缝图像受到各种环境的干扰,从实用角度出发,提出了结合局部灰度特征、局部相异度特征和局部方向特征的裂缝目标提取算法。首先以数量统计值对裂缝目标和背景粗划分,对粗划分后裂缝目标计算其灰度特征,结合当前点相异度特征并计算自适应阈值得到裂缝信息。但由于裂缝存在空洞点和断裂点,弥补方法是判断当前点邻域内非零像素点数,符合条件下再计算窗口内非零像素点与当前点的斜率,对角度进行投票,提取局部方向特征描述。最后结合多结构元素形态学去噪算子,去除孤立噪声。与最大类间方差法和基于传统特征的分割算法进行了对比,实验表明文中方法能够较为完整、连续地提取路面裂缝目标。A practical method for pavement crack object extraction based on Local Gray Feature（ LGF）,Local Dissimilarity Feature（ LDF） and Local Orientation Feature（ LOF） is proposed,aiming at extracting crack targets in various complex conditions. Firstly,rough division of crack target and background is done by the count statistics and gray feature is computed as LGF description,LGF- LDF feature is obtained for the crack basic information with adaptive thresholds,combined with the current point dissimilarity feature. To fill void and breaking points and to enhance computing efficiency,the number of non- zero pixels in the neighbor of the processing pixel point is judged and the orientation of the non- zero point and the processing point is computed to vote for the angle and the local orientation feature description is extracted. The morphological filter operators are utilized combing with multi- structure elements to eliminate isolated noises. The experiment results showthat the method proposed achieves a better performance in extracting the crack targets than both Otsu and algorithm based on traditional features.

关 键 词：路面裂缝 灰度特征 相异度特征 方向特征 特征描述 去噪 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]