prtools.register#

register(arr, ref, oversample, return_error=False)[source]#

Compute the subpixel image translation to register the input array to a reference array.

The registration shift is computed in two steps: first a coarse estimate is computed from the FFT-based cross-correlation of the two input arrays. This estimate is then refined to subpixel accuracy by computing the upsampled DFT-based cross-correlation in a small neigborhood around the initial estimate.

Parameters:

  • arr (array_like) – Array to register.

  • ref (array_like) – Target array.

  • oversample (float) – Oversampling factor for subpixel registration. Registration accuracy is 1/oversample.

  • return_error (bool, optional) – If True, the noramlized RMS registration error is returned. Default is False.

Returns:

  • shift (tuple) – Translation in (row, col) that will register arr to ref.

  • err (float) – Registration error

References

Guizar-Sicairos, Thurman, and Fienup, “Efficient subpixel image registration algorithms”. Optics Letters 33, 156-158 (2008)

See also

shift()

Example

>>> ref = np.zeros((3,3))
>>> ref[1,1] = 1
>>> arr = np.zeros((3,3))
>>> arr[2,2] = 1
>>> shift = prtools.register(arr, ref, oversample=2)
>>> shift
(-1.0, -1.0)