In [4]:
histogram(img_in, bins=2**8, interval=[0, 1])
Methods which provide color adjustment without input parameters.
With these techniques, all result is given by the auto analysis from the input image.
%matplotlib inline
import matplotlib as mpl
import matplotlib.pyplot as plt
import numpy as np
import imageio
from _utils import *
import warnings
warnings.filterwarnings('ignore')
img_in = imageio.imread('../_data/aerial01.png')/255\
histogram(img_in, bins=2**8, interval=[0, 1])
summary(img_in)
R G B
min: 0.0000 0.0000 0.3294
1st Quar: 0.4784 0.4314 0.6471
median: 0.6706 0.5569 0.7176
mean: 0.6283 0.5505 0.7126
3rd Quar: 0.7961 0.6824 0.7882
max: 0.9137 0.9098 0.9059
sigma: 0.1932 0.1693 0.0962
Automatic contrast enhancement using histogram equalization for each channel.
img_eq = img_in.copy()
# Split channels
R = img_eq[:,:,0]
G = img_eq[:,:,1]
B = img_eq[:,:,2]
# Get dimensions
N1, N2, _ = img_eq.shape
# Histogram equalization
Rs = np.sort(R.ravel())
Gs = np.sort(G.ravel())
Bs = np.sort(B.ravel())
R[:,:] = Rs.searchsorted(R)
G[:,:] = Gs.searchsorted(G)
B[:,:] = Bs.searchsorted(B)
# Feature normalization
img_eq = img_eq/(N1*N2 - 1)
histogram(img_eq, bins=2**8, interval=[0, 1])
summary(img_eq)
R G B
min: 0.0000 0.0000 0.0000
1st Quar: 0.2478 0.2480 0.2425
median: 0.4961 0.4990 0.4969
mean: 0.4967 0.4968 0.4943
3rd Quar: 0.7496 0.7432 0.7399
max: 1.0000 1.0000 1.0000
sigma: 0.2874 0.2885 0.2881
Feature scaling, using min-max normalization to stretch each channel to 0.0-1.0 range.
img_rgb = img_in.copy()
# Get minimum values
RGB_min = np.apply_over_axes(
np.amin, img_rgb, [0, 1]
).ravel()
# Get maximum values
RGB_max = np.apply_over_axes(
np.amax, img_rgb, [0, 1]
).ravel()
# min-max normalization
img_rgb = (img_rgb - RGB_min)/(RGB_max - RGB_min)
histogram(img_rgb, bins=2**8, interval=[0, 1])
summary(img_rgb)
R G B
min: 0.0000 0.0000 0.0000
1st Quar: 0.5236 0.4741 0.5510
median: 0.7339 0.6121 0.6735
mean: 0.6876 0.6050 0.6648
3rd Quar: 0.8712 0.7500 0.7959
max: 1.0000 1.0000 1.0000
sigma: 0.2115 0.1861 0.1668
Stretch components from HSV space to 0.0-1.0 range, only preserving the hue channel.
# Convert from RGB to HSV
HSV_ = mpl.colors.rgb_to_hsv(img_in)
# Get minimum values from HSV
HSV_min = np.apply_over_axes(
np.amin, HSV_, [0, 1]
).ravel()
# Get maximum values from HSV
HSV_max = np.apply_over_axes(
np.amax, HSV_, [0, 1]
).ravel()
# Find delta = max - min
HSV_delta = HSV_max - HSV_min
# Preserve hue component
HSV_min[[0]] = 0
HSV_delta[0] = 1
# Stretch the components, preserving hue
HSV_ = (HSV_ - HSV_min)/HSV_delta
# Convert from HSV to RGB
img_hsv = mpl.colors.hsv_to_rgb(HSV_)
histogram(img_hsv, bins=2**8, interval=[0, 1])
summary(img_hsv)
R G B
min: 0.0000 0.0000 0.0000
1st Quar: 0.3963 0.3624 0.5436
median: 0.6198 0.5154 0.6644
mean: 0.5936 0.5208 0.6600
3rd Quar: 0.7987 0.6886 0.7949
max: 1.0000 0.9933 0.9890
sigma: 0.2361 0.2109 0.1675
# Find the mean values
R_mu, G_mu, B_mu = np.apply_over_axes(
np.mean, img_in, [0, 1]
).ravel()
# Define the correction values based on green channel
corrR = G_mu - R_mu
corrB = G_mu - B_mu
# Offset the components to equalize all mean values
offset = img_in + [corrR, 0 ,corrB]
# Move the mean value to zero
cent = offset - G_mu
# Find the maximum values after offset
cent_max = np.apply_over_axes(
np.amax, cent, [0, 1]
).ravel()
# Stretch and normalize values gratter than mean
img_wb = (1 - G_mu)*cent/cent_max + G_mu
# Stretch and normalize values lower than mean
cent = offset + 1 - G_mu
cent_min = np.apply_over_axes(
np.amin, cent, [0, 1]
).ravel()
norm = 1 - (G_mu*(1 - cent)/(1 - cent_min) + 1 - G_mu)
mu_down = norm < G_mu
img_wb[mu_down] = norm[mu_down]
histogram(img_wb, bins=2**8, interval=[0, 1])
summary(img_wb)
R G B
min: 0.0000 0.0000 0.0000
1st Quar: 0.4192 0.4314 0.4563
median: 0.6171 0.5585 0.5621
mean: 0.6078 0.5681 0.5857
3rd Quar: 0.8147 0.7155 0.7263
max: 1.0000 1.0000 1.0000
sigma: 0.2257 0.1895 0.1785
print('Input')
histogram(img_in, bins=2**8, interval=[0, 1])
print('Equalize')
histogram(img_eq, bins=2**8, interval=[0, 1])
print('RGB component stretch')
histogram(img_rgb, bins=2**8, interval=[0, 1])
print('HSV component stretch')
histogram(img_hsv, bins=2**8, interval=[0, 1])
print('White balance')
histogram(img_wb, bins=2**8, interval=[0, 1])
Input
Equalize
RGB component stretch
HSV component stretch
White balance