Second-Order Statistical Texture Representation of Asphalt Pavement Distress Images Based on Local Binary Pattern in Spatial and Wavelet Domain

Journal of Rehabilitation in Civil Engineering 7-3 (2019) 48-67 Journal homepage: http://civiljournal.semnan.ac.ir/ Second-Order Statistical Texture Representation of Asphalt Pavement Distress Images Based on Local Binary Pattern in Spatial and Wavelet Domain R. Shahabian Moghaddam 1 , A. Mohammadzadeh Moghaddam 1 , S.A. Sahaf 1 and H.R. Pourreza 2 1. Department of Civil Engineering, Ferdowsi University of Mashhad, Mashhad, Iran. 2. Department of Computer Engineering, Ferdowsi University of Mashhad, Mashhad, Iran. Corresponding author: ARTICLE INFO Article history: Received: 10 March 2018 Accepted: 12 June 2018 Keywords: Pavement Distress Texture, Computer Vision, Gray Level Co-Occurrence Matrix (GLCM), Local Binary Pattern (LBP), Wavelet Transform. ABSTRACT Assessment of pavement distresses is one of the important parts of pavement management systems to adopt the most effective road maintenance strategy. In the last decade, extensive studies have been done to develop automated systems for pavement distress processing based on machine vision techniques. One of the most important structural components of computer vision is the feature extraction method. In most of the application areas of image processing, textural features provide more efficient information of image regions properties than other characteristics. In this research, three different algorithms were used to extract the feature vector and statistically analyzing the texture of six various types of asphalt pavement surface distresses. The first algorithm is based on the extraction of images second-order textural statistics utilizing gray level cooccurrence matrix in spatial domain. In second and third algorithms, the second-order descriptors of images local binary patterns were extracted in spatial and wavelet transform domain, respectively. The classification of the distress images based on a combination of K-nearest neighbor method and Mahalanobis distance, indicates that two stages arranging of the gray levels of the distress images edges by applying wavelet transform and local binary pattern (third algorithm) had a superior result in comparison with other algorithms in texture recognition and separation of pavement distresses. Classification performance accuracy of the distress images based on first, second and third feature extraction algorithms is 61%, 75% and 97%, respectively. DOI: 10.22075/JRCE.2018.14785.1272 R. Shahabian Moghaddam et al./ Journal of Rehabilitation in Civil Engineering 7-3 (2019) 48-67 1. Introduction In Iran, which more than 90% of freight and passenger transportation is relied on overland transport network, the road network has a pivotal role in country's wealth and must be fully capable of preserved [1]. To determine the apropos (economic) road treatment, the pavement management system (PMS) must be implemented. Pavement evaluation is one of the most significant elements of pavement management systems. Pavement assessment includes a range of qualitative and quantitative measurements to determine the operational and structural pavement conditions. Collecting the pavements assessment data is in terms of four sections including serviceability, structural capacity, surface failure (pavement distress) and safety (fraction). Identification and tracking of road surface distresses are considered as one of the main parts of the pavement performance assessment process at all levels of road management. In addition, one of the primary reasons for reducing road serviceability is pavement distresses [2]. Basically, manual (visual) methods are administrated to determine and measure the pavement distresses. Experiences has shown that this pavement scoring approach, despite its high accuracy, costs a considerable amount of time and money. It is also dependent on evaluator’s personal judgments (subjective) and will lead to unstable results. In the last decade, in order to fix the visual distress assessment defects, comprehensive researches has been done to develop semiautomated and fully-automatic methods of pavement inspection. In full-automated pavement assessment, all stages of gathering and processing distress date are done without or very few human intervention [3, 4]. 49 Chua and Xu [5] used the images moment invariants in spatial domain in order to sever and detect the distress area. Then they classified several types of asphalt pavement cracking by connecting discrete cracks and using an algorithm based on geometric properties and they reported more than 55% errors at the end. In the mentioned article, programming was utilized to identify the type of cracking. Writing an application that can describe a variety of distresses is associated with a lot of mistakes because of severe irregularities, lack of a precise definition of most distress patterns and the huge amount of important information behind the distress data. Therefore, implementing algorithms based on machine learning are preferable in analyzing pavement distress images. Nallamutho and Wang [6] utilized auto-correlation function to describe the asphalt cracks texture. The basis of this method is the measurement of interval alternation between the patterns. The texture can be described by auto-correlation function based on the intervals between the texels. In the aforementioned paper, the K-nearest neighbor (KNN) method was used to classify the distress images and the classification accuracy was about 56%. As noted, the values of the gray levels, which constitute the pavement surface distress textures, are of a random nature and very disorganized. So considering a specific locative relation between pixel's quantity (or texels) and interval alternation is not a truthful distress representing approach. Gray Level Cooccurrence Matrix (GLCM) (used in this article) uses two adjacent pixel's values to extract texture statistics. Chang et al. [7] used 11 texture features extracted from GLCM (Haralick texture descriptors) in spatial domain, to analyze the texture of asphalt pavement surface distress images. The 50 R. Shahabian Moghaddam et al./ Journal of Rehabilitation in Civil Engineering 7-3 (2019) 48-67 classification of these images based on decision tree learning method, resulted in about 35% errors. In the aforementioned research, in addition to using a large number of features and high computing loads, the GLCM has been formed unidirectional and symmetrically. Whiles, locating distribution of the distress gray levels should be analyzed in different directions by considering the order of pixel placement (non-symmetric GLCM) Lee [8] used the histogram statistical moments of the Fourier coefficients to analyze the texture of various types of pavement surface cracks. Afterwards he performed the support vector machine classifier to discriminate the images. The result of classification accuracy rate was about 72%. Fourier transform not only doesn’t provide sparse information of the distress images important information, s (...truncated)