Isodata

The isodata() function segments a :class:DataCube using the ISODATA algorithm—a variant of KMeans that iteratively splits and merges clusters based on spectral variance and cluster population.

Example

The following example illustrates the use of isodata() on a synthetically generated DataCube.

import wizard
from wizard._utils.example import generate_pattern_stack
from wizard._processing.cluster import isodata
import matplotlib.pyplot as plt

# generate synthetic data with spatial patterns
data = generate_pattern_stack(15, 300, 300, n_circles=4, n_rects=2, n_triangles=2)

# create a DataCube
dc = wizard.DataCube(data)

# apply ISODATA clustering
labels_iso = isodata(dc, k=5, it=25, p=4)

# visualize a wavelength slice and the segmentation
fig, axes = plt.subplots(1, 2, figsize=(10, 5))

axes[0].imshow(dc[10])
axes[0].set_title("Original Slice (λ index 10)")
axes[0].set_axis_off()

axes[1].imshow(labels_iso)
axes[1].set_title("Segmented (ISODATA)")
axes[1].set_axis_off()

plt.tight_layout()
plt.show()

The output shows how ISODATA adaptively determines cluster counts and captures both subtle and broad spectral patterns within spatially coherent regions.