LAUSR

The radiography testing (RT) method is one of the most common tools used for the inspection of industrial specimens for defect detection. Although modern industrial radiography systems are capable of providing radiography images with a high dynamic range, the high sensitivity of their receptor and unfavorable exposure conditions could lead to high levels of noise/fogginess on acquired images. It is therefore important to optimize the operator’s image perception and interpretation by improving the detection limit through suppression of the noise level. The development of effective image noise removal (denoising) algorithms remains an important research area. Therefore, in this study, three nonlocal-based regularization algorithms were used to improve RT defect detection capabilities. As a first step, a Tikhonov model was implemented to obtain the preprocessed image. Three different nonlocal regularization algorithms, L2-norm minimization (L2NM), H1 semi-norm (H1SN), and total variation regularization models, were then applied to achieve the required image enhancement. The methods relied on generating a regularized, smoothed image, which was then subtracted from the original image to reconstruct the enhanced final image for analysis. The results of this study on a set of RT images from the GDXray database showed that the three methods were capable of improving defect detection while preserving the object’s edge and fine detail imaging information satisfactorily. Comparison of the three methods showed that the L2NM method provided reconstructed images with the highest contrast in the defect regions.