Second-Order Statistical Texture Analysis of Asphalt Pavement Distresses Based on Local Binary Pattern in Spatial and Wavelet Domain

Assessment of pavement distresses plays a pivotal role in pavement management systems in determination of the most efficient option for repair and maintenance of the road. In the past decade, extensive researches have been made in order to develop automatic methods for pavement distresses processing based on computer vision techniques. Of the most important components comprising machine vision systems is the feature extraction procedure. In many image processing applications, textural features provide more effective information about the properties of the image regions in comparison with other features.  In the present study, three different algorithms were utilized with the purpose of statistically analyzing the textures of six different groups of asphalt pavement distress images. The first feature extraction algorithm is based on gray level co-occurrence matrix (GLCM) textural statistics. In second and third algorithms, the second-order textural descriptors of the images local patterns were extracted in spatial and wavelet domain, respectively. The distress categorization results based on a fusion of K-nearest neighbor (KNN) classifier and Mahalanobis distance, indicate that two level (double) regularizing the distress edges gray levels via employing wavelet transform and local binary pattern (third algorithm) outperforms other textural feature extraction algorithms in pavement distresses recognition and discrimination. The distress classification accuracy rate based on first, second and third algorithms were 61%, 75% and 97%, respectively.