Comparison of Methods for Processing of X-ray images of Defects in Reinforced Concrete Product

Abstract

Abstract The problem of automatic determination of the location of defect projections on radiographs of metal and reinforced concrete products is considered. The main research method is a computational experiment. A three-dimensional phantom has been developed that simulates a fragment of a concrete slab reinforced with iron bars with defects inside. Its X-ray image is modeled on the basis of the Bouguer law. Distortion represents uncorrelated noise with a normal distribution in each pixel. A binary Bayesian classifier is used to search for defects. It has been shown to be quite effective as long as the noise SDV does not exceed 1.5% of the average image brightness. At a higher noise level, the classifier does not give stable results. The use of simple low-frequency filtering methods (averaging and median in a sliding window) for noise suppression did not lead to improvement. However, the use of the entropy filter has shown that it can improve the quality of classification. Special image point detectors, in particular the Harris-Stephans detector, were also used to search for defects. The results obtained suggest that this approach may be promising.