Detection of Pavement Cracks Based on Non-local Image Denoising and Enhancement

Yingkun Hou, Hao Hou, Guang-Hai Liu, Junjie Hou

Abstract: Non-local method is a very effective method to eliminate Gaussian white noise in images, but the pavement image noise is usually non-Gaussian. In this paper, a non-local based method of denoising and enhancement of pavement crack images is proposed. First, morphological corrosion operator is used to convert the image noise of pavement into near-Gaussian noise; Second, using an improved (Block-match 3D Filtering) BM3D denoising and enhancement method to denoise and enhance pavement images; Finally, the remaining noise is removed by block cosine transform combined with gradient value. The experimental results show that the proposed method can achieve ideal pavement cracks detection results.