Skip to content

Registration Sync Task

The Registration sync task computes a translational alignment between frames using frequency-domain cross-correlation and applies the computed subpixel shift to reproject the input frame. The task uses a masked, mean-centered cross-correlation pipeline implemented with FFTs:

  1. center the mean of the input (optionally restricted by an ellipse mask derived from radius),
  2. compute forward FFTs (R2C) of the input and stored reference,
  3. form the normalized cross-power spectrum (conjugate × product, normalize magnitude),
  4. inverse FFT to obtain the cross-correlation surface,
  5. find the integer peak and refine it with a 3×3 subpixel estimator,
  6. apply the computed subpixel shifts to the input using bilinear interpolation (wrap-around).

Typical uses include translational registration of single-frame images, subpixel alignment prior to averaging, and motion correction pipelines.

Warning

The radius setting must be within [0, 1]. Values at the extremes may produce degenerate masks (e.g. radius == 0 selects nothing, radius == 1 selects the full frame). Choose a radius appropriate to the region of interest for robust cross-correlation.

Inputs

This task consumes a single input tensor.

The input requirements (as enforced by RegistrationFactory::infer and the implementation) are:

  • The tensor must have at least 2 dimensions (height and width).
  • The dtype must be float32 (F32).
  • The tensor must reside in device (GPU) memory.
  • The task expects a single frame (batch size 1) for proper operation; non-unit batches are not processed by the algorithm.

Outputs

This task produces one output tensor with the same shape and dtype as the input (the pipeline computes shifts and writes the shifted frame).

The memory location of the output tensor is device memory.

Inplace

This task does not support inplace operation.

Ownership

This task does not own any inputs or outputs.

Settings

Example:

1
2
3
{
    "radius": 0.8
}
Property Pattern Type Deprecated Definition Title/Description
+ radius No number No - Normalized radius of the circular mask applied to the region of interest. Defines the elliptical area used for cross-correlation computation. Must satisfy:
- 0.0 ≤ radius ≤ 1.0
- Radius is normalized relative to the smallest image dimension