Study Of Image Enhancement For Pavement Crack Based On The Partial Differential Equations

Crack detection has been a difficult problem to solve in disease automatic detection, andhas become the main research interest of this thesis. In order to protect the crack informationwhile remove noise in the pavement crack image, this thesis proposed a new method toremove noise in pavement crack image using weighted model of improved P-M model andthe coherent enhancement diffusion. This model overcome the disadvantage of P-M equation,Coherent Enhancement Diffusion and other basic partial differential equation models, whichcannot noise reduction and protect crack while strong interference of complicated noises andsundries exists in the pavement images. The new diffusion model encourage the CoherentEnhancement Diffusion and inhibit P-M diffusion, and in the non-crack section that meansflat area of image, the new diffusion model encourage P-M diffusion and inhibit coherentenhancement diffusion. This method can enhance crack flow structure and sharpen crack edgewhile denoising. Through this way, we realize preprocessing of crack image’s earlier stage.After denoising and enhanced the crack image we binary it based on piece, this proposedmethod overcomes the disadvantage of the global threshold value method, the globalthreshold value method doesn’t consider the space correlation of each pixels in the image, andaccording to the uniform standard select threshold. This proposed method can separate thetarget cracks from the background better. But use this method in the target cracks which areweak in noise environment, they can not be detected. Therefore, we use valley point detectionto detect the target cracks. But there also exist pseudo crack. Lastly, the pavement crackextraction and pseudo crack elimination are realized with the method of using mathematicalmorphology to process the binary image, thus we obtain the pavement crack detection results.The simulation results show the proposed method in this thesis can remove oil stain,pebble and other impurities interference from pavement crack images, and pseudo crack. Butthis crack detection method still needs to be improved to overcome the case of few isolatednoise pixels, and interrupted crack edges.