基于小尺度分形维数的裂缝图像分割方法

A Crack Image Segmentation Method Based on Small Scale Fractal Dimension

提出了一种基于小尺度分形维数理论的裂缝图像分割方法。首先使用中值滤波处理,将要处理像素点的某个邻域内的所有像素按照灰度值的大小进行排列,取排列的中值作为该像素点的新值,进而让周围的像素接近真实值,从而消除孤立的噪声点,在保留裂缝信息的基础上能有效去除噪声的影响。其次,根据提出的小尺度分形维数算法,将维数细化到最小值使其图像的细节得到最大的体现。比较传统的分形维数方法,该方法可以使裂缝的不规则性得到更好的体现,能够对裂缝的不规则特征进行分割,可以得到更好的效果。通过计算路面图像中每个像素点的分形维数,实现了对裂缝信息的有效提取。最后通过大量试验并借助准确率-召回率评价体系验证了所提方法的有效性。根据不同算法的裂缝图像分割性能比较可知:传统的方法在准确率上并不能有较好的结果。通过曲线图可以看出:传统的方法与提出的方法仍然存在着一定的差距。在F-测度上提出的方法基本都稳定在0.9左右,而传统方法的F-测度大多都徘徊在0.7以下,二值化数据可以下降到0.3以下。试验结果表明:提出的方法能够有效实现道路图像的裂缝图像分割。

A crack image segmentation method based on the theory of small scale fractal dimension is proposed. First,the median filtering is applied to arrange all pixels in a neighborhood of pixel point to be processed according to the size of the gray value. The median value of the arrangement is used as a new value of the pixel point to make the surrounding pixels close to the true value,thereby eliminating the isolated noise points,and the effect of noise can be effectively removed on the basis of preserving crack information.Second,according to the proposed algorithm of small scale fractal dimension,the dimension is refined to the minimum to maximize the image details. Compared with the traditional fractal dimension method,this method can better reflect the irregularity of the cracks,can divide the irregular characteristics of the cracks,and can get better effect. By calculating the fractal dimension of each pixel point in road surface image,the crack information can be effectively extracted. Finally,in order to the effectiveness of the proposed method is verified by a large number of test and precision-recall evaluation system. According to the comparison of crack image segmentation performance of different algorithms,traditional method cannot achieve satisfactory result in accuracy rate. It can be seen from the curves that there is still a certain gap between traditionalmethod and the proposed method,the F-measures of the proposed are basically stable at around 0. 9,while most of the F-measures of traditional methods are below 0. 7,and binaryzation data can fall below 0. 3. The experimental result shows that the proposed method can achieve satisfying result of crack image segmentation.