View : 12 Download: 0

Fast and accurate marker-based projective registration method for uncalibrated transmission electron microscope tilt series

Title
Fast and accurate marker-based projective registration method for uncalibrated transmission electron microscope tilt series
Authors
Lee H.Lee J.Shin Y.G.Lee R.Xing L.
Ewha Authors
이레나
SCOPUS Author ID
이레나scopus
Issue Date
2010
Journal Title
Physics in Medicine and Biology
ISSN
0031-9155JCR Link
Citation
vol. 55, no. 12, pp. 3417 - 3440
Indexed
SCI; SCIE; SCOPUS WOS scopus
Abstract
This paper presents a fast and accurate marker-based automatic registration technique for aligning uncalibrated projections taken from a transmission electron microscope (TEM) with different tilt angles and orientations. Most of the existing TEM image alignment methods estimate the similarity between images using the projection model with least-squares metric and guess alignment parameters by computationally expensive nonlinear optimization schemes. Approaches based on the least-squares metric which is sensitive to outliers may cause misalignment since automatic tracking methods, though reliable, can produce a few incorrect trajectories due to a large number of marker points. To decrease the influence of outliers, we propose a robust similarity measure using the projection model with a Gaussian weighting function. This function is very effective in suppressing outliers that are far from correct trajectories and thus provides a more robust metric. In addition, we suggest a fast search strategy based on the non-gradient Powell's multidimensional optimization scheme to speed up optimization as only meaningful parameters are considered during iterative projection model estimation. Experimental results show that our method brings more accurate alignment with less computational cost compared to conventional automatic alignment methods. © 2010 Institute of Physics and Engineering in Medicine.
DOI
10.1088/0031-9155/55/12/010
Appears in Collections:
의학전문대학원 > 의학과 > Journal papers
Files in This Item:
There are no files associated with this item.
Export
RIS (EndNote)
XLS (Excel)
XML


qrcode

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.

BROWSE