Based On Curvelet Transform And LBP Operator For The Research Of Pavement Crack Identification

Once there are some crack diseases on bituminous pavements, it will become heavier in an increasing traffic and serious overloading.Then will greatly shorten the life of the asphalt pavement. Due to this condition, which needs to do Real Time Detection on road, and find defect problem on time. This paper mainly in-depth study and research image preprocessing, the filter de-noising, Curvelet transform and Local Binary Pattern(LBP) and other related image processing technology, apply it to the road image, and then a detailed analysis of the asphalt pavement rapid detection and identification technologies, and proposed a new asphalt pavement recognition algorithm.The main work is as below:(1) Research illumination compensation model based on LBP operator. Collected asphalt pavement image inevitably has a complex background texture, noise pollution, taking into account the characteristics of LBP under intense light, noise or complex imaging conditions classified extreme performance degradation problems, proposed using histogram equalization performed illumination compensation model, and do the experiment contrast, further analysis of the method under the enhanced image preprocessing result, the difference in the image can be improved with the road surface cracks background.(2) Proposed a new model of Canny-HBT. HBT filter is proposed by Saravana Kuma, the edge similarity measure of HBT filter could weak frequency component information, strengthen image edges, lines of high frequency information. But the detection method, in the final image edge detection and location, does not have sufficient structural similarity. There are some gaps on the canny edge detection operator on the third criterion of the minimum response. So the paper uses CannyHBT filter to modify some parameters,which could reanch to thin edge of the secondary treatment, the result proved new Canny-HBT filter can get better denoising effect.(3) Proposed a new feature fusion approach based on curvelet transform and multi-scale LBP for disease recognition. Firstly, the image of light on the road after pretreatment disease is divided into blocks, and its different sub-blocks of different scales were used to extract LBP operator histogram feature vector. Link up each subblock of The LBP histogram features as pavement crack image’s LBP feature vector; then do decomposition based on pavement cracks images. Curvelet feature vectors extracted by modifying the optimal scale layers and layers of fine-scale factor, and finally LBP features and characteristics of different scales Curvelet were in accordance with guidelines of high / low-frequency coefficients were weighted fusion coefficient fusion to generate a feature space for classification. In this paper, using image that is processed by Canny-HBT filter as as an input image of the algorithm,and using simulation to verify the proposed algorithm,compared to single LBP or Curvelet algorithm has a high recognition rate.