Image Processing Of Pavement Crack And Its Application

This thesis mainly studies the methods used in processing asphalt pavement’s image-crack automatic recognition, and solves some problems that might arise in image-crack automatic recognition, such as the uneven grayscale, the lower contrast, the various and complex noise, the shadow, the uncertain threshold, the unrecognized cracks and so on, which has a very important significance.This thesis includes the following aspects:(1) For pavement image with lower grayscale, less details and larger dynamic areas, the method based on wavelet transformation and PCNN model (pulse coupled neural networks), together with the image-strengthened algorithm by using improved Retinex, is put forward to enhance the image. The former, combining with PCNN model’s various activating means, makes use of wavelet decomposition to strengthen the gradient of the image, effectively highlighting the cracks. The latter, according to Retinex theory and the detailed information of gradient and image, constructs a composite smooth conduction function, avoiding the bad effects caused by the single-scale Retinex method and the large amount of calculation of the multi-scale method. So it highlights the crack information, keeps the details, and plays a good enhancement.(2)For pavement image with complex noise, image-noise-reduction method for the three types of noise is designed. For Gaussian noise, noise processing method based on transformed domain and estimated noise is adopted. For impulse noise and salt-pepper noise, based on the characteristics of the noise, improved median filtering is used. And experimental results show that the three kinds of algorithm can effectively remove the noise and at the same time better retain details.(3)For pavement image with uneven grayscale, the algorithm by correcting the uneven illumination based on the statistical properties and the grayscale of the sampling window is put forward. The algorithm, combining with the characteristics of the normal distribution of the pavement image as well as uneven imaging principle, makes use of the block thinking and helps solve the problem of uneven image grayscale. The experimental results show that the algorithm is able to achieve the purpose of correcting uneven illumination.(4)For pavement image which can not use simple threshold segmentation, the hierarchical threshold-crack-segmentation method is proposed, based on statistical model. According to the characteristics of the normal distribution of the pavement image, dual-threshold is determined and the skeleton-point information of cracks is extracted. A single threshold is used to obtain the approximate shape-pixel information of cracks, and then the crack information is reestablished by combining with regional expansion method and skeleton points. Finally, mathematical morphology is used to repair cracks, thus achieving the accurate segmentation of pavement image.(5)For binary image of the pavement image, the five characteristics of projection in both X-axis and Y-axis are extracted. The application of fuzzy neural network approach is to realize the classification and identification of cracks image, which avoids the shortcomings of GA and SOM classified methods under two characteristics and achieves a better classification results.The methods mentioned in the thesis are theoretically validated and analyzed in detail, and a lot of experiments are done to prove the effectiveness of the algorithm.