Learning Gabor filters with scikit-learn and ICA or k-means

learning_gabor_filters.py

import numpy as np
import matplotlib.pyplot as plt

from sklearn.datasets import fetch_mldata
from sklearn.decomposition import FastICA, PCA
from sklearn.cluster import KMeans

# fetch natural image patches
image_patches = fetch_mldata("natural scenes data")
X = image_patches.data

# 1000 patches a 32x32
# not so much data, reshape to 16000 patches a 8x8
X = X.reshape(1000, 4, 8, 4, 8)
X = np.rollaxis(X, 3, 2).reshape(-1, 8 * 8)

# perform ICA
ica = FastICA(n_components=49)
ica.fit(X)
filters_ica = ica.unmixing_matrix_

# zero mean "by hand" so that inverse transform
# doesn't mess up the filters
X -= X.mean(axis=0)
# perform whitening
pca = PCA(n_components=49, whiten=True)
X_white = pca.fit_transform(X)
kmeans = KMeans(k=49, n_init=1).fit(X_white)

filters_kmeans = pca.inverse_transform(kmeans.cluster_centers_)
filters_pca = pca.components_

titles = ["ICA", "PCA", "k-means"]
filters = [filters_ica, filters_pca, filters_kmeans]

for T, F in zip(titles, filters):
    plt.figure(T)
    for i, f in enumerate(F):
        plt.subplot(7, 7, i + 1)
        plt.imshow(f.reshape(8, 8), cmap="gray")
        plt.axis("off")
plt.show()