In [4]:
histogram(sample, interval=[0, 1])
Implementation of solution to image stacking and statistical blending.
%matplotlib inline
import matplotlib.pyplot as plt
import matplotlib.gridspec as gridspec
import imageio
import numpy as np
import scipy.stats as stats
import glob
from _utils import *
sample = imageio.imread('../_data/SEQ01-32bits/sample.0001.exr')
print('sample 0001')
summary(sample)
sample 0001
R G B
min: 0.0294 0.0204 0.0148
1st Quar: 0.1513 0.1422 0.1306
median: 0.2340 0.2164 0.1900
mean: 0.3025 0.2680 0.2191
3rd Quar: 0.3858 0.3417 0.2785
max: 71.7715 47.6547 24.3402
sigma: 0.3659 0.2595 0.1554
histogram(sample, interval=[0, 1])
Transform from linear to sRGB usign gamma correction $\large \gamma = 2.2$.
sample_sRGB = sample**(1/2.2)
histogram(sample_sRGB, interval=[0, 1])
def stackRead(pathname):
'''
pathname defined by "glob" pattern.
i.e.: "directory/sequence_folder/image_*.jpg"
'''
# List of image in pathname folder
SEQ_IMG = glob.glob(pathname)
n = len(SEQ_IMG)
# sample for stack definition
sample = imageio.imread(SEQ_IMG[0])
# x and y are the dimensions
# c is the number of channels
y, x, c = sample.shape
# define stack
stack = np.zeros((n, y, x, c), dtype=sample.dtype)
# image stacking
for FILE in SEQ_IMG:
index = SEQ_IMG.index(FILE)
stack[index] = imageio.imread(FILE)
# output
return stack
stack = stackRead('../_data/SEQ01-32bits/sample.*.exr')
panel(stack**(1/2.2), (3, 1),
interval=[0, 1],
dims=(1200, 400),
texts=['{:04}'.format(i + 1) for i in range(10)])
def blendStack(stack, modo='median', axis=0):
if modo == 'sum':
blend = np.sum(stack, axis)
if modo == 'arithmetic mean':
blend = np.mean(stack, axis)
if modo == 'geometric mean':
blend = stats.gmean(stack, axis)
if modo == 'harmonic mean':
blend = stats.hmean(stack, axis)
if modo == 'median':
blend = np.median(stack, axis)
if modo == 'minimum':
blend = np.amin(stack, axis)
if modo == 'maximum':
blend = np.amax(stack, axis)
if modo == 'curtosis':
blend = stats.kurtosis(stack, axis)
if modo == 'variance':
blend = np.var(stack, axis)
if modo == 'standard deviation':
blend = np.std(stack, axis)
return blend.astype(stack.dtype)
median = blendStack(stack)
summary(median)
R G B
min: 0.0344 0.0230 0.0167
1st Quar: 0.1516 0.1433 0.1319
median: 0.2307 0.2143 0.1851
mean: 0.2911 0.2606 0.2128
3rd Quar: 0.3766 0.3335 0.2688
max: 2.4678 1.6451 0.8355
sigma: 0.2001 0.1651 0.1150
histogram(median**(1/2.2), interval=[0, 1])
sample_blend = np.array([stack[0]**(1/2.2), median**(1/2.2)])
panel(sample_blend, (2, 1),
interval=[0, 1],
texts=['sample 0001', 'median'])
blend = blendStack(stack**(1/2.2), modo='sum')
summary(blend)
R G B
min: 6.8831 5.8026 4.9726
1st Quar: 13.7512 13.3464 12.8770
median: 16.6397 16.0336 15.0072
mean: 17.5535 16.7477 15.4227
3rd Quar: 20.7067 19.5605 17.7030
max: 58.8951 49.4278 36.5980
sigma: 5.0968 4.5682 3.7053
histogram(blend, interval=[5, 38])
sample_blend = np.array([stack[0]**(1/2.2), (blend - 5)/(38 - 5)])
panel(sample_blend, (2, 1),
interval=[0, 1],
texts=['sample 0001', 'sum'])
mean_a = blendStack(stack**(1/2.2), modo='arithmetic mean')
mean_g = blendStack(stack**(1/2.2), modo='geometric mean')
mean_h = blendStack(stack**(1/2.2), modo='harmonic mean')
sample_blend = np.array([mean_a, mean_g, mean_h])
panel(sample_blend, (3, 1),
dims=(1200, 400),
interval=[0, 1],
texts=['arithmetic mean', 'geometric mean', 'harmonic mean'])
minimum = blendStack(stack**(1/2.2), modo='minimum')
maximum = blendStack(stack**(1/2.2), modo='maximum')
sample_blend = np.array([minimum, maximum])
panel(sample_blend, (2, 1),
interval=[0, 1],
texts=['minimum', 'maximum'])
