Machine Learning for OpenCV

Reducing the color palette using k-means

Now let's reduce these 16 million colors to just 16 colors by telling k-means to cluster all color variations into 16 distinct clusters. We use the before mentioned procedure, but now specify 16 as the number of clusters:

In [9]: criteria = (cv2.TERM_CRITERIA_EPS + cv2.TERM_CRITERIA_MAX_ITER,...                 10, 1.0)...     flags = cv2.KMEANS_RANDOM_CENTERS...     img_data = img_data.astype(np.float32)...     compactness, labels, centers = cv2.kmeans(img_data,...                                               16, None, criteria,...                                               10, flags)

The resulting cluster corresponds to the 16 colors of our reduced color palette. Visual inspection of the centers array reveals that all colors have three entries--B, G, and R--with values between 0 and 1:

In [10]: centersOut[10]: ...