Skip to content

Align2d

Jump to bottom
Adrian Quintana edited this page Dec 11, 2017 · 1 revision

Align2d

Purpose

This utility allows you to align translationally and rotationally a set of 2D-images. If no reference is provided an intial one is constructed by piramidally cross-correlating a sub-population of N=2^k images of the input selfile, where N is the lower nearest value to the total number of images in the input selfileMarco et al., 1996, Ultramicroscopy 66. By default, alignment of the images is performed using the algorithm proposed byPenczek et al., 1990, Ultramicroscopy 40, employing alternative translational and rotational alignments. In the first iteration all particles will be aligned with respect to the reference image (provided by-ref, or obtained by piramidal cross-correlation). For all subsequent iterations, the current image is subtracted from the reference before alignment, and added again afterwards.

An alternative refinement scheme is provided by the-complete option that performs a complete 3-dimensional search of the rotation angle and both translations. Although much slower, it may be more robust than the default option, especially in the case of non-spherically shaped particles and/or low signal-to-noise ratios. However, since the rotational angle is usually not sampled very finely, and only integer shifts are possible, the accuracy of this option is rather low. Furthermore, the reference has been observed to shift somewhat upon iterative application of the -complete option. It is therefore recommended to finish such a refinement with manual replacement of the reference in the center, and some additional iterations using the conventional scheme of alternating translational and rotational alignments.

The output transformations are not applied to the matrices, but stored in the header of the images. (Therefore, as long as you save the header information of the images in document files using the-doc option of this program, or alternatively the headerinfo program, you can savely overwrite the input images. The headerinfo program can then be used to restore any angles/shifts obtained previously.) Besides the individual images, also the average and standard deviation images of the aligned particles are written out as:[rootname]_[extension].med.xmp and[rootname]_[extension].sig.xmp. If rotational alignment is performed, a file called[rootname]_[extension].corr will be written out that shows the average correlation coefficient of all aligned images for all angles. This file can be helpful to determine the symmetry of the particles. During the refinement the current reference image is (over)written as[rootname].ref.xmp after each iteration. Usingxmipp_show -img rootname.ref.xmp -poll allows to follow the refinement process on-line. If piramidal combination of images is used to construct the initial reference, this image is written out as:[rootname].pspc.xmp. It is centered with respect to the average of all non-aligned particles.

Usage


$ align2d  -i [rootname.sel] ...

Parameters

  • `` Input selfile with images to be aligned
  • `` Reference image (if no reference is provided an initial one will be constructed by piramidally cross-correlating a sub-population of the images)
  • `` Extension for output images[filename].[extension], output selfile`[rootname][extension].sel` and output average & standard deviation images of the successfully aligned particles`[rootname][extension].med.xmp` and`[rootname]_[extension].sig.xmp`. If none is provided input images and selfile will be overwritten
  • `` Write document file with rotations and translations.Strongly recommended
  • -iter [int1] = Number of refinement iterations to perform
  • -Ri [inner radius0] =
  • -Ro [outer radiusdim/2] = Only region between radii Ri and Ro [in pixels] will be considered for rotational correlation
  • `` Fourier-filter images to expected resolution (in Angstroms)
  • `` Sampling rate, i.e. pixel size (in Angstroms), needed to apply Fourier-filter
  • `` Discard images which shift more in the last iteration (in pixels)
  • `` Discard images which rotate more in the last iteration (in degrees)
  • `` Skip rotational alignment, only perform translational one
  • `` Skip translational alignment, only perform rotational one
  • `` Use complete 3-dimensional search of the rotation angle and both translations for alignment
  • -psi_step [float10] = For complete-search: sampling interval of the rotation angle (in degrees)

Examples and notes

A typical 2D-alignment process could look like this:

I. Pre-center images with respect to an initial average image of rotationally non-aligned particles (g0u.med.xmp, calculated using statis)


$ align2d -i g0u.sel -ref g0u.med.xmp -only_trans -doc precenter.doc

I. Translational & rotational alignment of all images using Penczeks algorithm, obtaining the initial reference using a piramidal combination of a subset of the input images.


$ align2d -i g0u.sel -filter 10 -sampling 2 -Ri 10 -Ro 25 -max_shift 2 -max_rot 3 -doc process.doc

The input images and selfile are overwritten. The average and standard deviation images after the last iteration are written as:g0u.med.xmp andg0u.sig.xmp. After each refinement iteration the current reference image is written asg0u.ref.xmp. By typingxmipp_show -img g0u.ref.xmp -poll in another window you can follow the refinement process on-line. The reference image resulting from the piramidal combination of images is written as:g0u.pspc.xmp. The average correlation coefficient for all angles is written in:g0u.corr

  • It is always a good idea to output document files like precenter.doc and process.doc. Since these allow to restore the angles and shifts from a previous point in the alignment process, using the program headerinfo

Non-spherically shaped particles with a low signal-to-noise ratio may be better aligned using an alignment scheme like this:

I. Pre-center images with respect to an initial average image of rotationally non-aligned particles (g0u.med.xmp, calculated using statis):


$ align2d -i g0u.sel -ref g0u.med.xmp -only_trans -doc precenter.doc -iter 10

II. Initial alignment of all images using a complete search of the translational and rotation parameters, obtaining the initial reference using a piramidal combination of a subset of the input images.


$ align2d -i g0u.sel -filter 10 -sampling 2 -Ri 10 -Ro 25 -doc complete.doc -complete -psi_step 5

III. After checking the center of the reference, further refinement using Penczeks algorithm:


$ align2d -i g0u.sel -ref g0u.med.xmp -filter 10 -sampling 2 -Ri 10 -Ro 25 -doc refine.doc -iter 10

USER's COMMENTS

--Main.AlfredoSolano - 17 Jan 2007

Clone this wiki locally