How to ignore a color or alpha when using clusters

Question

I am trying to find the dominant color of an image using Pil and cluster. My problem is that my images has a transparent background because these are .png and so i always get black as the dominant color. I'd like to ignore the first dominant color and pick the second most dominant color.

Is there a way to ignore alpha color or just remove it from the result? I am afraid that by just removing the first most dominant color, i would sometimes remove the actual dominant color in case of the background being a really small part of the image.

Here is my code :

from PIL import Image
import numpy
import math
import matplotlib.pyplot as plot
from sklearn.cluster import MiniBatchKMeans

imgfile = Image.open("images/abra.png")
numarray = numpy.array(imgfile.getdata(), numpy.uint8)

X = []
Y = []

fig, axes = plot.subplots(nrows=5, ncols=2, figsize=(20,25))

xaxis = 0
yaxis = 0

cluster_count = 3

clusters = MiniBatchKMeans(n_clusters = cluster_count)
clusters.fit(numarray)

npbins = numpy.arange(0, cluster_count + 1)
histogram = numpy.histogram(clusters.labels_, bins=npbins)
labels = numpy.unique(clusters.labels_)

barlist = axes[xaxis, yaxis].bar(labels, histogram[0])
if(yaxis == 0):
    yaxis = 1
else:
    xaxis = xaxis + 1
    yaxis = 0
for i in range(cluster_count):
    barlist[i].set_color('#%02x%02x%02x' % (
    math.ceil(clusters.cluster_centers_[i][0]),
        math.ceil(clusters.cluster_centers_[i][1]), 
    math.ceil(clusters.cluster_centers_[i][2])))


plot.show()

Here is en example of my current code :

Image given :

Returned values :

Answer 1

You could avoid passing transparent pixels into the classifier like this, if that's what you mean:

#!/usr/bin/env python3

from PIL import Image
import numpy as np
import math
import matplotlib.pyplot as plot
from sklearn.cluster import MiniBatchKMeans

# Open image
imgfile = Image.open("abra.png")

# Only pass through non-transparent pixels, i.e. those where A!=0 in the RGBA quad
na = np.array([f for f in imgfile.getdata() if f[3] !=0], np.uint8)

X = []
Y = []

fig, axes = plot.subplots(nrows=5, ncols=2, figsize=(20,25))

xaxis = 0
yaxis = 0

cluster_count = 3

clusters = MiniBatchKMeans(n_clusters = cluster_count)
clusters.fit(na)

npbins = np.arange(0, cluster_count + 1)
histogram = np.histogram(clusters.labels_, bins=npbins)
labels = np.unique(clusters.labels_)

barlist = axes[xaxis, yaxis].bar(labels, histogram[0])
if(yaxis == 0):
    yaxis = 1
else:
    xaxis = xaxis + 1
    yaxis = 0
for i in range(cluster_count):
    barlist[i].set_color('#%02x%02x%02x' % (
    math.ceil(clusters.cluster_centers_[i][0]),
        math.ceil(clusters.cluster_centers_[i][1]), 
    math.ceil(clusters.cluster_centers_[i][2])))


plot.show()