Content-type: text/html Manpage of blindDeconvolution


Section: User Commands (1)
Index Return to Main Contents


blindDeconvolution - refine a subvolume using blind deconvolution  


blindDeconvolution volName
blindDeconvolution volName psfSize
blindDeconvolution volName psfSize nIters
blindDeconvolution volName psfSize nIters saveFreq  


Attempt to refine a subvolume average using the blind deconvolution algorithm of Hirsch et al (J. Computational Biology 3:335-346, 2011) with sparseness priors. The observed volume is assumed to result from blurring of the true volume by convolution with an unknown kernel of specified maximum size, plus added noise. I.e. we assume y = f*x, where the input volume is a potentially noisy version of y.

Estimates of f, x, and y will be refined iteratively and the final versions saved to <volName>_<psfSize>_<iterNum> with suffixes of .psf, .x, and .y, respectively. Often the "sharpened" *.x volume will the desired output. The *.y volume, while not sharpened, can also be useful as a de-noised version of the input. Cross sections of both the kernel, y, and x are displayed during refinement, along with plots of the log of the posterior probability density, and the residual sum of squared errors.

Both the kernel and true volume are required to be non-negative. If the input volume contains any negative values, an offset will be added to eliminate them before processing. Blind deconvolution is an extremely ill-conditioned process. Varying kernel size and the number of iterations will typically lead to different results. Until external confirmation is obtained, the most that one can say about a given result is that it is consistent with the observed data.

is the path to the MRC file containing the input subvolume. The volume should be padded to be constant within at least psfSizevoxels of any edge.
is a single, odd number giving the width of the cubical kernel in voxels. Initially, all voxels in the kernel will be set to the same value. If ommitted, a default of ~1/16th of the input volume size will be used. Kernel size can significantly alter the results; trying kernel sizes of 5, 7, and 9 voxels is often a good starting point.
is the maximum number of refinement iterations to execute. Output files will be written after completing the specified number of iterations (Default = 100).
is an optional integer specifying that output files should also be written after every <saveFreq> iterations (Default = 10).


John Heumann  






This document was created by man2html, using the manual pages.
Time: 18:16:05 GMT, January 11, 2021