Python中图像像素强度与色彩度量

Question

I want to measure the average pixel intensity and measure of colourfulness of a image. For this i'm following this approach (kindly let me know if there's any alternative approach for the same):

a) Calculate Average Pixel Intensity:

im = Image.open('images-16.jpeg')
stat = ImageStat.Stat(im)
r,g,b = stat.mean
mean = sqrt(0.241* (r ** 2) + 0.691* (g ** 2) + 0.068* (b ** 2))
print(mean)

b) To measure colourfulness:

• Dividing color space into 64 cubic blocks with four equal partitions along each dimension

   w,h=im.size bw,bh = 8, 8 #block size img = np.array(im) sz = img.itemsize shape = (h-bh+1, w-bw+1, bh, bw) strides = (w*sz, sz, w*sz, sz) blocks = np.lib.stride_tricks.as_strided(img, shape=shape, strides=strides) print (blocks[1,1]) 

• Calculate Euclidean distances between the geometric centers C i of each cube i Not able to compute (say d(a,b)=rgb(Ca)-rgb(Cb))

• Distribution D1 is generated as the color distribution of a hypothetical image such that for each of 64 sample points, the frequency is 1/64 -

  pixels = im.load() all_pixels = [] for x in range(218): #put your block width size for y in range(218): #your block heigh size cpixel = pixels[x, y] all_pixels.append(cpixel)

• Distribution D2 is computed from the given image by finding the frequency of occurrence of color within each of the 64 cubes How can i do this?

• Calculate Earth Mover's Distance: (D1,D2,d(a,b)) - d(a,b) is calculated above

Is this the right way to do it? Any supporting documents to achieve this? Any help with the code is appreciated. Thanks.

Answer 1

You will need pyemd library.

from pyemd import emd
import numpy as np
from PIL import Image
import skimage.color

im = Image.open("t4.jpg")
pix = im.load()

h1 = [1.0/64] * 64
h2 = [0.0] * 64
hist1 = np.array(h1)

w,h = im.size

for x in xrange(w):
    for y in xrange(h):
        cbin = pix[x,y][0]/64*16 + pix[x,y][1]/64*4 + pix[x,y][2]/64
        h2[cbin]+=1
hist2 = np.array(h2)/w/h

# compute center of cubes

c = np.zeros((64,3))
for i in xrange(64):
    b = (i%4) * 64 + 32
    g = (i%16/4) * 64 + 32
    r = (i/16) * 64 + 32
    c[i]=(r,g,b)

c_luv = skimage.color.rgb2luv(c.reshape(8,8,3)).reshape(64,3)

d = np.zeros((64,64))

for x in xrange(64):
    d[x,x]=0
    for y in xrange(x):
        dist = np.sqrt( np.square(c_luv[x,0]-c_luv[y,0]) + 
                   np.square(c_luv[x,1]-c_luv[y,1]) + 
                   np.square(c_luv[x,2]-c_luv[y,2]))
        d[x,y] = dist
        d[y,x] = dist


colorfullness = emd(hist1, hist2, d)

print colorfullness