def spectral(image_data: np.ndarray) -> np.ndarray:
"""
Run Spectral Clustering on given image.
Resources
* https://sklearn.org/auto_examples/cluster/plot_face_segmentation.html#sphx-glr-auto-examples-cluster-plot-face-segmentation-py
* https://sklearn.org/auto_examples/cluster/plot_segmentation_toy.html#sphx-glr-auto-examples-cluster-plot-segmentation-toy-py
"""
scale = 0.3
scaled_size = (
int(image_data.shape[1] * scale), int(image_data.shape[0] * scale)
)
# image_data_scaled is (m * scale, n * scale, 3)
image_data_scaled = cv2.resize(image_data, scaled_size)
# reduced is (m * scale * n * scale, 1)
reduced = flatten_image(image_data_scaled)
# labels is (m * scale * n * scale, 1)
labels = SpectralClustering(
n_clusters=2, eigen_solver="amg", assign_labels="discretize",
n_jobs=-1
).fit_predict(reduced)
labels.resize(scaled_size[1], scaled_size[0])
# label_mask is (m * scale, n * scale, 1)
label_mask = (get_foreground_label(labels) * 255).astype("uint8")
prediction = apply_prediction(label_mask, image_data_scaled)
prediction_upscale = cv2.resize(
prediction,
(image_data.shape[1], image_data.shape[0])
)
return prediction_upscale
