Statistics Properties of Asphalt Pavement Images for Cracks Detection

In order to overcome the illumination interference on the asphalt pavement cracks detection, the higherorder statistics based non-uniformity illumination correction method is firstly proposed, which improves the image quality significantly. The local regions of pavement have similar geometric texture, and then the probability distribution of image pixel values in local regions is also similar. Based on the fact a new method to calculate statistical correlation characteristics of pixel values in local region is adopted. And then the feature vector is built by joint the pixel value and the characteristics values of statistical correlation. Support vector machine is used for rough detection by training the sample feature vectors and prediction processing. Finally morphological dilation, corrosion, connectivity region label and area constraints are adopted to remove impurities and complete the finer cracks detection. The experimental results show that compared with the detection method based on Otsu image segmentation, the detection accuracy is significantly improved, the proposed method can not only detect the cracks with weak contrast but also detect the tiny cracks.