Fast Template Matching via Cross-Correlation

Share:

Description

Motion correction and/or co-registration of three-dimensional arrays (medical imaging data) are performed by applying a user-defined mask of voxels. Normalized cross-correlations (in 3D) are computed using the FFT.

Usage

1
2
## S4 method for signature 'array'
ftm(input, ...)

Arguments

input

is a four-dimensional array of signal intensities.

...

Additional variables passed to the plot function.

Details

An extremely basic method of motion correction/co-registration is implemented by estimating “local” cross-correlations based on a binary mask that is a subset of the original three-dimensional volume. All convolutions are preformed via the FFT (fft) and repetitive calculations are minimized where possible.

Only whole-voxel translations are considered. This does not begin to capture the true effects of motion in soft tissue, but we assume that the object of interest (e.g., tumor) is a fairly rigid structure. Potential extensions include rigid-body, affine and nonlinear registration techniques along with interploation schemes in order to capture intra-voxel manipulations of the data.

Value

A list of objects are returned:

out

Motion-corrected version of the four-dimensional array.

offset

Translations (in 3D) for each volume in the 4D array.

t.center

Estimated center of the binary mask.

Author(s)

Brandon Whitcher <bjw34032@users.sourceforge.net>

References

Lewis, J.P. (2003) Fast normalized cross-correlation.
www.idiom.com/~zilla/

See Also

conv.fft, find.center, shift3D