Cracks are considered one of the most critical problems of pavement systems. Fully automatic distress evaluation systems are designed to assess pavement conditions. They are developed by combining image processing and pattern recognition techniques for detecting and classifying roadway surface distresses. Currently, there is a considerable need for developing fully automatic distress evaluation methods that are robust, fast and cost-effective and apply to any surface regardless of the pavement texture or type.

In this thesis, two automatic crack assessment methods are proposed for the detection and classification of cracks from acquired 2-D pavement images. In the first method, a region-based image processing algorithm was developed. First, local thresholding was applied, and then the existence of cracks inside each region was detected based on spatial distribution of crack pixels. By fitting curves in the center of regions, they were connected and their type and length were determined by measuring the slopes and lengths of the curves. In the second method, a pixel-based image processing method was developed that uses an improved version of the weighted neighborhood pixels segmentation algorithm for crack detection purposes. This method uses the Gaussian cumulative density function as the adaptive threshold to obviate the problem of using fixed thresholds in noisy environments. The developed methods were tested on 100 and 300 images of Asphalt Concrete (AC) and Portland Cement Concrete (PCC) surfaces respectively; the validation results were measured for the first method as: Precision = 0.85, Accuracy = 0.87, Recall = 0.89 and F1 score =0.87 and for the second method as Precision = 0.79, Accuracy = 0.99, Recall = 0.95, and F1 score = 0.86. The validation results were compared with the detection results of some fully automatic methods in the literature, and their competencies were verified.