Industrial computed tomography for three-dimensional cultural relic model reconstruction based on L1-αL2+TV norm minimization

Abstract

Industrial computed tomography (ICT) has emerged as a valuable tool for the conservation of cultural relics since it enables the extraction of information about the internal structure of artifacts without causing any damage. However, sparse and finite angle projection computed tomography (CT) often generates images with a large number of artifacts caused by insufficient data. This paper introduces a new cone-beam CT hybrid model based on compressed sensing (CS) theory and total variance (TV) denoising using the Split Bregman algorithm to address this problem, for the first time using compressed sensing theory for CT reconstruction of cultural relics. The model combines the L1 and L2 norms and optimizes the total variance of the image as the objective function. To enhance the reconstruction of L1-L2, the paper employs the continuation strategy of α and proposes a new combination function to update α. Consequently, the reconstruction results are compared with those of traditional common methods, and the effect of different α update functions is observed on the reconstruction results. The results from the model and real CT images demonstrate that the proposed L1-αL2 + TV method yields superior reconstruction results compared to ART + TV, SART, and L1/L2 algorithms based on objective evaluation indexes such as RMSE, SSIM, and PSNR, the proposed L1-αL2 + TV method achieves SSIM improvement of at least 15.2 %. Moreover, the newly proposed combined function is better at preserving the grain of the image, resulting in a much-improved reconstruction of the contours, making it highly beneficial for subsequent conservation and restoration of heritage. Overall, this proposed L1-αL2 + TV mixture model provides low-noise and high-quality reconstruction results, while maintaining the grain of the original image, making it a great boon for heritage conservation and restoration.