Enhancing flaw detection in aluminum castings by two different mixed noise removal methods

Abstract

Aluminum casting is utilized for making complex objects. Different defects may, however, be introduced during the casting process; the detection of which relies on radiography testing as a standard inspection method. To improve information extraction from the acquired X-ray images, image processing methods are often necessary to improve the contrast of the image features and to increase detection success of hidden defects. In this study, two methods based on the sparse regularization were used to enhance the contrast and defect(s) visualization from the radiographs of different casting objects. The Weighted Encoding with Sparse Nonlocal Regularization (WESNR) and Laplacian Scale Mixture (LSM) with a Nonlocal Low-rank Regularizer (NLR) was used to remove Gaussian and impulse noises from the low contrast images. The proposed algorithms were successfully implemented to radiographic images of the cast objects. The results show that improvements in the visualization of internal structure and defect regions were significant, usually by factors of between two and three. Quantitative data supported the outcome of radiography operators’ evaluation that in general, the image features from reconstructed images using the LSM-NLR reconstructed images were better visualized than the WESNR enhanced images.