Compare image filters

Apply three different 3D filters to the same CT image: a 5-voxel Mean filter, a 1.5 mm Laplacian-of-Gaussian (LoG) filter, and a first-level Daubechies 3 wavelet LLH response. Save each response and compare its mean and standard deviation. Mean smoothing retains the CT intensity scale; LoG and wavelet values are transformed responses rather than raw HU.

import numpy as np

from zrad.filtering import LoG, Mean, Wavelets3D
from zrad.image import Image

image = Image.from_nifti("path/to/phantom.nii.gz")

# Use the same input and boundary handling for each filter.
filters = {
    "mean": Mean(padding_type="reflect", support=5, dimensionality="3D"),
    "log": LoG(padding_type="reflect", sigma_mm=1.5, cutoff=4.0, dimensionality="3D"),
    "wavelet_llh": Wavelets3D(
        wavelet_type="db3", padding_type="reflect",
        response_map="LLH", decomposition_level=1,
    ),
}

for name, image_filter in filters.items():
    response = image_filter.apply(image)
    response.save_as_nifti(f"output/{name}.nii.gz")
    # These summary statistics cover the full image, including outside the ROI.
    print(f"{name}: mean={np.mean(response.array):.3f}, std={np.std(response.array):.3f}")

The files mean.nii.gz, log.nii.gz, and wavelet_llh.nii.gz can be opened in an image viewer to inspect spatial differences. See Python filtering for other filters and parameters.