Using Interpolation with Nonlocal Autoregressive Modeling for Defect Detection in Welded Objects

X-rays and gamma rays are used in industrial radiography to detect internal defects and different structures of the test object. The radiography interpreters must be able to evaluate and interpret the radiography images as accurately as possible. To improve the operator’s image perception and interpretation, the quality of radiographs can be enhanced by different image processing methods. In this study, the sparse representation method with a nonlocal autoregressive model (NAM) based on a sparse representation model (SRM) algorithm was implemented to improve the defect detection capabilities. The technique relies on generating a regularized and smoothed image, which is then subtracted from the original image to reconstruct the high contrast image. The algorithm was successfully applied to different radiography images. Improved defect detection was achieved while preserving the fine details and the main information of the images. For the enhanced images of samples in this study, figures of merits were found between 83 and 98% for the different defects and regions of interests in the reconstructed radiographs by the NAM–SRM algorithms. These figures of merit were between 67 and 89% in the original radiographs, respectively. The results show that the reconstructed images by NARM–SRM algorithms have better visualization and also the defect regions are very clear to the original radiographs. Regarding computing time, the proposed method is faster than the other four chosen iterative methods.