This notebook reduces the number of colors using scikit learn KMeans algo.
- Each RGB color is a 3-dimension vector where each dimension can have 256 values (from 0 to 255 included).
- The KMeans clustering algorithm is used on 10% of randomly selected pixels.
- Each original pixel is then mapped to the closest cluster center.
- The resulting image has a (much) smaller number of colors.
- The number of colors necessary for the human eye to be insensitive to any additional color granularity is what we want to discover here.
- Because integers are coded as bits, the interesting number of colors are the powers of two: 2, 4, 8, 16, 32, 64, 128, 256, etc.
- But it is not immediate to go from color reduction to smaller file size as we can see from the file sizes.
The new IPywidgets are used to see the result. Very convenient.