Extraction of pavement cracks based on valley edge detection of fractional integral

Abstract

As pavement crack images are difficult to segment due to the existence of high noise, weak boundary and small cracks, an extraction method of pavement cracks based on the valley edge detection of fractional integral is proposed. In this method, first, neighboring smoothing of the original image is performed to eliminate the noise and expand the relative width of the cracks. Then, the main cracks are extracted via the valley edge detection of fractional integral, and the resulting image is further processed via the morphological approach with short-line noise elimination. Afterwards, final cracks are extracted by using the gap linking method on maximum entropy threshold to cause cracks to merge automatically. Experimental results show that the proposed method instantly helps to detect small pavement cracks with high accuracy and strong noise robustness.