Demo about effect of the different stiching kernels

%matplotlib inline
import matplotlib.pyplot as plt

import scipy.misc
import numpy
from nabu.testutils import get_data

raw_data_1 = get_data('brain_phantom.npz')['data']
raw_data_2 = numpy.random.rand(*raw_data_1.shape) *255.0

from nabu.stitching.overlap import OverlapStitchingStrategy, ZStichOverlapKernel

plot the two raw data side by side

fig,ax = plt.subplots(2)
ax[0].imshow(raw_data_1)
ax[1].imshow(raw_data_2)

fig = plt.figure()
plt.title("raw data")
plt.subplot(121)
plt.imshow(raw_data_1)
plt.subplot(122)
plt.imshow(raw_data_2)

def plot_strategy_result(overlap_strategy, img_1, img_2, with_weights=True, extra_params=None):
    strategy = OverlapStitchingStrategy.from_value(overlap_strategy)
    kernel = ZStichOverlapKernel(
        frame_width=img_1.shape[1],
        stitching_strategy=strategy,
        overlap_size=img_1.shape[0],
        extra_params=extra_params,
    )
    stitch = kernel.stitch(img_1, img_2)[0]
    if with_weights:
        w_1 = kernel.weights_img_1
        w_2 = kernel.weights_img_2
    else:
        w_1 = w_2 = None

    fig = plt.figure(constrained_layout=True)
    if with_weights:
        axes = fig.subplot_mosaic(
            [
                ["00", "01"],
                ["10", "11"],
                ["2", "2"],
            ],
        )
    else:
        axes = fig.subplot_mosaic(
            [
                ["2", "2"],
            ],
        )
    if w_1 is not None and w_2 is not None:
        plt.title(f"{strategy.value} strategy")
        # plot weights curves
        axes["00"].set_title("image 1 weigths")
        axes["00"].plot(numpy.arange(len(w_1)), w_1, color="red")
        axes["01"].set_title("image 2 weigths")
        axes["01"].plot(numpy.arange(len(w_2)), w_2, color="blue")

        # plot weighted images
        axes["10"].set_title("image 1 with weights applied")
        axes["10"].imshow(img_1*w_1)
        axes["11"].set_title("image 2 with weights applied")
        axes["11"].imshow(img_2*w_2)

    # plot stitched images
    axes["2"].set_title("stitched image")
    axes["2"].imshow(stitch)

mean strategy

plot_strategy_result(
    overlap_strategy="mean",
    img_1=raw_data_1,
    img_2=raw_data_2,
)

closest strategy

plot_strategy_result(
    overlap_strategy="closest",
    img_1=raw_data_1,
    img_2=raw_data_2,
)

linear weights

plot_strategy_result(
    overlap_strategy="linear weights",
    img_1=raw_data_1,
    img_2=raw_data_2,
)

cosinus weights

plot_strategy_result(
    overlap_strategy="cosinus weights",
    img_1=raw_data_1,
    img_2=raw_data_2,
)

image minimum divergence

This method will not applied a simple predetermined weights on the two images.

It split the two images into two parts: high frequency and low frequency.

The two low frequency parts will be stitched using a ‘sinusoidal’ / cosinus weights approach. When the two high frequency part will be stitched by taking the lower divergent pixels

plot_strategy_result(
    overlap_strategy="image minimum divergence",
    img_1=raw_data_1,
    img_2=raw_data_2,
    with_weights=False,
    extra_params={
        "frequency_threshold": 5,
    },
)
# uncomment to compare vs 'raw' sinusoidal
# plot_strategy_result(
#     overlap_strategy="cosinus weights",
#     img_1=raw_data_1,
#     img_2=raw_data_2,
#     with_weights=False,
# )

higher signal

This method will not applied predetermine weights on the two images.

It will pick pixel on the image having the higher signal.

A use case is that if there is some artefacts on images which creates stripes (from scintillator artefacts for example) it could be removed from this method

raw_data = get_data("brain_phantom.npz")["data"]
new_dataset_1 = raw_data_1.copy()
new_dataset_1[40:75] = 0.0
new_dataset_1[:, 210:245] = 0.0

new_dataset_2 = raw_data_1.copy()
new_dataset_2[:, 100:120] = 0.0

fig = plt.figure()
plt.title("raw data")
plt.subplot(121)
plt.imshow(new_dataset_1)
plt.subplot(122)
plt.imshow(new_dataset_2)

plot_strategy_result(
    overlap_strategy="higher signal",
    img_1=new_dataset_1,
    img_2=new_dataset_2,
    with_weights=False,
)