Recognition Method Of Pavement Distress Images Fusing Manifold Features

With the rapid development of computer and digital image technology, the image of autumatic identififcation techonology based on digital images of the road surface is widely used in road repair work and good results have been achieved. However there are still some technical difficulties in this method need to be resolved. For example the denoising effect is not ideal, the fracture features of image enhancement effect is not obivous and the image low-level visual features such as damage density factors, coordinate projections could not effectively distinguish the different types of pavement cracks. In order to overcome the above problems, in the paper the dictionary learning techniques is applied to remove noise and enhance the fracture characteristics, and the manifold learning algorithm is used to analyse the high-dimensional pavement image data dimension reduction, extracted a low dimensional manifold features and feature optimization of pavement cracks. The main research work in the paper is as follows:(1) Analyse and compare the KLLD algorithm, the algorithm of median fifter and histogram equalization algorithm of damaged image denoising effect, the expermental results show that the KLLD alogorithm is the best.(2) An adaptive algorithm for image denoising and enhancement was presented. Firstly, the algorithm, we need estimated the noise of the road gray image, If the image noise is greater than a given threshold, then using KLLD alogorithm to deal with the image.then estimate the strength of the image crack characteristics, if the crack characteristics estimated value is less than the given threshold,further using the KLLD algorithm to enhance the denoising image.(3) A recognition method fusing manifold features was presented. In the method, a low dimensional manifold features firstly were extracted from pavement images with Laplacian Eigenmaps as semantic feature. Then manifold features were fused with distress density factors or coordinate projections. The categories of cracks were recognized on the fused features. Experimental results showed that manifold features could effectively distinguish various pavement cracks and the recognition accuracy was greatly improved through fusing manifold features and other features.(4) The visualization of pavement distress image was achieved by using the manifold features. By analyizing the distributions of the two-dimensional manifold features, the physical meaning of two features was obtained. One dimension denotes the direction of cracks and another dimension describes the degree of pavement damage.