Pavement crack images typically have the characteristics of uneven distribution of illumination, strong noises, and a small proportion of cracks. Differentiating the cracks from the background image with the traditional grayscale analysis and edge detection methods is difficult. To solve this problem, an algorithm based on multi-scale shape analysis was developed. The algorithm divided a whole pavement image into 256 image cells. For every cell, an optimal threshold was selected to become a binary image. Under small scales conditions, 6 shape-factors were extracted from the binary cell image which composed a feature vector that was inputted into a BP neural network to classify the binary cells into two types: crack or non-crack. Lastly, under the conditions of big scales, the "wild spots" were deleted, and the cracks were located precisely on the whole image. Experimental results show that the proposed algorithm has an advantage in computing speed and accuracy rates.
展开▼
