In a variety of road diseases, much attention has been paid to cracks which often cause significant engineering and economic problems. The pavement images are usually characterized by the presence of multiple possible objects (such as oil spots and dirt, signs of lane, vegetation fragments and other non-pavement images), thus it is necessary to distinguish cracks from cluttered background, which increases the difficulty of crack detection. In order to solve the problems above, the MinMax k-Means algorithm based on prior knowledge is proposed firstly for crack detection. In the proposed method, clusters are assigned weights that can be self corrected in direct proportion to their variances in clustering process and the prior knowledge is introduced to deal with the problem that the clustering results are sensitive to the initial position of clustering centers. In addition, the pretreatment uses a method with a mean value of a cracked image block that is smaller than the mean value without cracks to pre-mark the image block and scan the mean matrix from both vertical and horizontal directions, which improve the accuracy of the clustering results greatly. The proposed algorithm is compared with the standard k-Means algo-rithm on a same sample and the comparison results show the accuracy and validity of the pro-posed algorithm.%在多种公路病害中,经常造成重大工程事故和经济问题的裂缝引起较大关注.公路路面图像通常包含各种物体,例如油渣、污垢、车道标记、植被碎片和其他非路面伪像,因此需要从杂乱的背景中区分裂缝,增加裂缝检测难度.为解决上述问题,提出基于先验知识的MinMax k-Means算法进行裂缝检测.该算法在聚类过程中分配与簇内方差大小成正比的可自动修正的权重,并引入先验知识以处理聚类结果对聚类中心初始位置敏感问题.除此之外,预处理采用含裂缝图像块的均值比不含裂缝的均值小的方法预标记图像块,并从垂直和水平两个方向扫描均值矩阵,大大提升聚类结果的准确性.使用相同样本,将所提算法与标准k-Means算法比较,可知所提算法有更好的准确性和有效性.
展开▼
