Registration

The registration process step consists of aligning the images with each other.

This process occurs in two steps.

• the calculation of the 3x3 (affine) transformation matrices associated to each pairs of consecutive images (based on the pystackreg library).

• the transformation step as the application of the previously calculated matrices to the full stack.

Illustration of the registration process step in the synthetic test case. Reminder: shifted slices are located in [20-70]

[registration_calculation]
#area = [5, 35, 5, 35]
threshold = 0.5
nb_blocks = [1, 1]
transformation = "TRANSLATION"

[registration_transformation]
#constant_drift = [-0.3, 1.2]
#box_size_averaging = 100
subpixel = false
mode = "edge"
cropping = true

area is an optional keyword used to calculate the transformation matrices with cropped images in the aim to reduce the computing time. Note that the area should be sufficiently large to encompass “features” relevant for the registration processing.

threshold is employed to binarize the images during the registration_calculation processing. If a threshold value is not specified (commented line), the registration will be performed based on the gray levels of the images.

nb_blocks allows decomposing the image into patches (according to a grid defined by the number of blocks in the x and y directions). This spatial division of the images thus enables accounting for non-uniform alignment effects on large images that could not be captured by a single transformation matrix.

transformation sets the type of the transformation to be considered (to choose among ‘TRANSLATION’, ‘RIGID_BODY’, ‘SCALED_ROTATION’, ‘AFFINE’, ‘BILINEAR’). See the pystackreg documentation for more details.

During the image acquisition, it is not uncommon for drift to occur (visible in the ‘cumulative’ transformation matrix resulting from the calculation step). constant_drift allows the user to remove the effect of a constant drift between each slide according to the values defined for each of the x and y directions.

In the case of a large-scale fluctuating drift, the user has the possibility to subtract the resulting mean component from a running average of a given size (number of slices) defined by the box_size_averaging parameter.

subpixel is an activation key to consider (or not) translation less or equal than 1 pixel. Figures below illustrate an application with subpixel = false.

mode is associated to the extrapolation method to be considered when applying the transformation matrices (to choose among ‘constant’, ‘edge’, ‘symmetric’, ‘reflect’, ‘wrap’ - see numpy.pad for more details. Note that if ‘constant’ is selected, the corresponding value (cval) associated to the extrapolation is by default NaN (np.nan)).

cropping is an activation key to perform cropping on the entire stack according to the area of aligned images preserved from any extrapolation effects (Default value is false).

Plotting

The special plotting related to the registration_calculation and the registration_transformation process steps generates images in the 2 dedicated outputs folders that are named registration_calculation.png and registration_calculation_cumul.png.

registration_calculation.png shows the transformation matrices values resulting from the registration of images pairs.

registration_calculation_cumul.png shows the corresponding cumulative transformation used for the full stack images realignment.

Note that setting subpixel = false in this use case makes the cumulative transformation flat (=no translation) outside the shift area [30-70] despite low non-zero translation values calculated during the registration_calculation process step.