Research On Pavement Crack Image Segmentation Method Based On Fractal Dimension

In recent years, the road network is gradually expanding. The development ofhighways has been steady-going, and the road maintenance and management issuesare becoming prominent. Crack is the initial form of the road pavement damages, andthe most important indicator to the quality of the road. There are many disadvantagesof the traditional manual testing method: inaccurate, low efficiency, high cost,time-consuming, impact on traffic, safety hazards, etc. Fast, accurate and efficientautomated pavement crack detection technology is the current development direction,and it is significant to the development of road transport undertakings.Fractal geometry is an emerging discipline, but it develops rapidly,. With thedevelopment of computer technology, the digital image processing technology hasbeen widely used. In this paper the differential box-counting (DBC) dimensionmethod has been improved, and combining with digital image processing technology,a new pavement crack image segmentation method based on the modified DBCmethod is proposed. First the original color image is made into the grayscale image.According to the gray projection curves of horizontal and vertical directions, the typeof the crack is determined, including the transverse crack, the longitudinal crack andthe reticulate crack, and then the image is shrank. Then the image is enhanced by thelogarithmic transform. After that the image is filtered using the improved medianfiltering method. Then the local fractal dimension (LFD) values of the image arecalculated using the improved differential box-counting method. In this paper, themethod dynamically overlaying the mesh is used to solve the limiting image size ofthe original method, and the idea grouping different scales is used to improve thespeed of the algorithm. Since the fractal dimension can describe different naturalobjects or textures, finally the optimal threshold of the LFD values is selected by thelocal dynamic threshold method, and then the LFD values are divided into twocategories, so the segmentation is done.In the end, several crack images of concrete pavement and asphalt pavement are selected to experiment using this method, and they are segmented by the classicalimage segmentation methods for comparison. It is proved that segmentation results ofthis method are better, with more accurate crack location, and better anti-noiseperformance.