A Novel adaptive PCA Based Denoising Technique for Ultra-High-Rate Computed Tomography

Publication Type:

Conference Paper


7th Conference on Industrial Computed Tomography (iCT 2017), ndt.net, Leuven (Belgium) (2017)


denoising, eigenvalue, principal component analysis (PCA), ultra-fast scan


Increasing the X-ray tomography acquisition rate is of high importance especially in medical applications as well as dynamic processes investigation. Thanks to the high brilliance of synchrotron radiation, it is possible to reduce the exposure time and do a tomographic scan with a sub-second temporal resolution which allows following dynamic processes in 4D (3D space + time).
On the other hand, increasing the acquisition rate leads to more background noise which strictly limits the advantages of high rate scan. We apply a new fast denoising technique using universal properties of eigen-spectrum of random covariance matrices. Our proposed technique is established based on the principal component analysis (PCA) of redundant data which shows that most of the signal-related variance is contained in a few components, whereas the noise is spread over all components. Extensive numerical evaluations of the proposed technique on a real dataset were acquired at the TOMCAT beamline with its ultra-fast endstation, show significant improvement in the quality of reconstructed images and elimination of noise.

Research area: