Crack Detection for Noisy Images Based on Combination of Likelihood Ratio Test and Principal Component Analysis

摘要

This paper presents a detection algorithm for cracks in noisy images of concrete structures by means of likelihood ratio test and principal component analysis (PCA). In this research, assuming that the cracks can be regarded as a straight line in the microscopic sense, the local gradient of the candidates for cracks is estimated by the PCA. In order to extract the crack shapes we employ the likelihood ratio test using a two-dimensional test function having the uniform profile of the cross-section along the estimated tangential line. Artificial long length straight lines included in the detected candidates for the cracks will be eliminated using the Hough transform.