Image reconstruction algorithm regularized by anisotropic gradient sparsity and low rank for cone beam computed...

Area of Science:

• Materials Science
• Imaging Science
• Computer Vision

Background:

• X-ray computed laminography (CL) is a key nondestructive imaging technique for plate-shell components.
• CL's unique geometry causes incomplete projection data, leading to aliasing artifacts and image degradation.
• Existing reconstruction methods struggle to fully address these CL-specific artifacts.

Purpose of the Study:

• To develop a novel CL reconstruction model addressing aliasing artifacts and structural degradation.
• To leverage both local smoothness and global spatial correlation in CL images.
• To improve the quality of reconstructed CL images for better component inspection.

Main Methods:

• Proposed an anisotropic gradient, sparse, and low-rank (AGSLR) regularization model for CL reconstruction.
• Incorporated total variation (TV) for local smoothness and tensor nuclear norm (TNN) for global spatial correlation.
• Introduced anisotropic TV terms to capture differential edge restoration and utilized the Chambolle-Pock (CP) algorithm for solving.

Main Results:

• The AGSLR model effectively suppresses cone-beam artifacts inherent in CL imaging.
• Demonstrated significant noise reduction in reconstructed CL images.
• Showcased superior recovery of structural details compared to existing methods.
• Validated performance on both simulated and real experimental data.

Conclusions:

• The proposed AGSLR regularization model offers a significant advancement in CL image reconstruction.
• This method enhances the reliability and accuracy of nondestructive inspection for plate-shell components.
• The AGSLR model provides a robust solution for overcoming CL's inherent imaging challenges.