Comparing Non-linear Diffusion and Total Variation Noise Reduction in Radiography Images of Welds

Industrial inspection of components is important for quality control procedures. Casting components and welded joints are regularly tested to see if they show irregularities or not. These irregularities (e.g., vertical and horizontal cracks) can extend through the part and cause catastrophic damages if they are not detected in the early stages of development. Several image processing and auto-detection schemes has been proposed so far, that are the combination of image processing techniques, and machine learning algorithms. In all of these systems, a preprocessing block is determined before further processing the image. In this step, noise reduction and image enhancement methods are utilized. In the previous studies, the low-pass filtering has been the method of choice, which is not flexible enough to handle the task-specific problems that may occur. In this paper, we have examined two famous noise reduction algorithms – Alvarez-Mazorra (AM) and total variation (TV) regularization – to compare their performance. This comparison is done according to each algorithm’s run-time, Peak Signal-to-Noise Ratio (PSNR) and Mean Squared Error (MSE). We have also represented the intensity plot of the central line in the weld bead to have a visualization of each method’s performance. Based on our results, the two algorithms have slight differences in their performance, but TV is preferred to AM algorithm because of less time cost.