Publication: A bayesian reconstruction method with marginalized uncertainty model for camera motion in microrotation imaging
Issued Date
2010-07-01
Resource Type
ISSN
00189294
Other identifier(s)
2-s2.0-77953787283
Rights
Mahidol University
Rights Holder(s)
SCOPUS
Bibliographic Citation
IEEE Transactions on Biomedical Engineering. Vol.57, No.7 (2010), 1719-1728
Suggested Citation
Danai Laksameethanasan, Sami S. Brandt A bayesian reconstruction method with marginalized uncertainty model for camera motion in microrotation imaging. IEEE Transactions on Biomedical Engineering. Vol.57, No.7 (2010), 1719-1728. doi:10.1109/TBME.2010.2043674 Retrieved from: https://repository.li.mahidol.ac.th/handle/20.500.14594/29082
Research Projects
Organizational Units
Authors
Journal Issue
Thesis
Title
A bayesian reconstruction method with marginalized uncertainty model for camera motion in microrotation imaging
Author(s)
Other Contributor(s)
Abstract
Reconstruction of a 3-D structure from multiple projection images requires prior knowledge of projection directions or camera motion parameters that describe the relative positions and orientations of 3-D structure with respect to the camera. These parameters can be estimated using, for instance, the conventional correlation alignment and feature-based methods. However, the alignment methods are not perfect, where the inaccuracy of the estimated motion parameters causes artifacts in the reconstruction. To overcome this problem, we propose a Bayesian approach to reconstruct the object that takes the motion uncertainty distribution into account. Moreover, we consider the motion parameters as nuisance parameters and integrate them out from the posterior distribution, assuming a Gaussian uncertainty model, which yields a statistical cost function to be minimized. The proposed method is applied in microrotation fluorescence imaging, where we aim at 3-D reconstruction of a rotating object from an image series, acquired by an optical microscope. The experiments with simulated and real microrotation datasets demonstrate that the proposed method provides visually and numerically better results than the traditional reconstruction methods, which ignore the uncertainty of the motion estimates. © 2006 IEEE.