Python Image Libraries

---

• Author: Diego Inácio
• GitHub: github.com/diegoinacio
• Notebook: image_libraries.ipynb

---

Brief overview about some of the main python libraries which promote input and output of digital image files.

%matplotlib inline
import matplotlib.pyplot as plt
import numpy as np

from _utils import *

import warnings
warnings.simplefilter('ignore', DeprecationWarning)

PIL

---

Reading and writing images using Python Image Library PIL.

Installation

---

Install by command line:

>>> conda install --channel anaconda pillow

or

>>> pip install Pillow

Input

---

PIL still doesn't support bit depth greater than 8.

from PIL import Image as pil

%%time
try:
    img_pil = pil.open('../_data/lenna.png')
    img_pil = np.asarray(img_pil)
except:
    print('Can not load image!')

Wall time: 66 ms

Analysis

---

summary(img_pil)

                          R               G               B
       min:         54.0000          1.0000          8.0000
  1st Quar:        146.0000         59.0000         78.0000
    median:        197.0000         97.0000        100.0000
      mean:        180.2240         99.0565        105.4126
  3rd Quar:        220.0000        135.0000        125.0000
       max:        255.0000        248.0000        239.0000
     sigma:         49.0486         52.8771         34.0609

%%time
histogram(img_pil)

Output

---

img = pil.fromarray(img_pil)
#img.save('_output/image_pil.tiff')

Wall time: 1.29 s

Scipy

---

Reading and writing images using SciPy.

Installation

---

Install by command line:

>>> conda install scipy

ou

>>> pip install scipy

Input

---

sciPy still doesn't support bit depth greater than 8.

import scipy.misc as scipy

%%time
try:
    img_scipy = scipy.imread('../_data/woman01.png')
except:
    print('Can not load image!')

Wall time: 13 ms

Analysis

---

summary(img_scipy)

                          R               G               B
       min:          9.0000          0.0000          6.0000
  1st Quar:         67.0000         56.0000         90.0000
    median:        123.0000        101.0000        123.0000
      mean:        129.2053         99.2666        125.1805
  3rd Quar:        173.0000        134.0000        150.0000
       max:        255.0000        255.0000        255.0000
     sigma:         62.3956         53.3887         39.1241

%%time
histogram(img_scipy)

Wall time: 1.38 s

Output

---

#scipy.imsave('_output/image_scipy.tiff', img_scipy8)

OpenCV

---

Reading and writing images using OpenCV.

Installation

---

Install by command line:

>>> conda install --channel menpo opencv

or

>>> pip install opencv-python

Input

---

openCV supports bit depth greater than 8. The result is in BGR mode and it's necessary invert channels to convert to RGB.

import cv2

%%time
try:
    img_cv = cv2.imread('../_data/woman02.png', -1)
    b, g, r = cv2.split(img_cv)
    img_cv = cv2.merge((r, g, b))
except:
    print('Can not load image!')

Wall time: 25 ms

Analysis

---

summary(img_cv)

                          R               G               B
       min:          0.0000         22.0000         34.0000
  1st Quar:        128.0000        139.0000        140.0000
    median:        145.0000        142.0000        148.0000
      mean:        137.5893        139.9451        144.0028
  3rd Quar:        147.0000        144.0000        151.0000
       max:        255.0000        255.0000        255.0000
     sigma:         33.8707         29.0540         28.2876

%%time
histogram(img_cv)

Wall time: 1.36 s

Output

---

#cv2.imwrite('_output/image_cv.tiff', img_cv)

Images - 16 and 32 bits

---

%%time
try:
    img_cv8 = cv2.imread('../_data/sphere_8bits.tiff', -1)
    b, g, r = cv2.split(img_cv8)
    img_cv8 = cv2.merge((r, g, b))
except:
    print('Can not load 8 bits image!')

try:
    img_cv16 = cv2.imread('../_data/sphere_16bits.tiff', -1)
    b, g, r = cv2.split(img_cv16)
    img_cv16 = cv2.merge((r, g, b))
except:
    print('Can not load 16 bits image!')

try:
    img_cv32 = cv2.imread('../_data/sphere_32bits.tiff', -1)
    b, g, r = cv2.split(img_cv32)
    img_cv32 = cv2.merge((r, g, b))
except:
    print('Can not load 32 bits image!')

Wall time: 121 ms

print('8 bits')
summary(img_cv8)

8 bits
                          R               G               B
       min:          3.0000          4.0000          3.0000
  1st Quar:         93.0000         92.0000         93.0000
    median:        117.0000        116.0000        118.0000
      mean:        129.5133        122.3053        129.6245
  3rd Quar:        164.0000        141.0000        164.0000
       max:        254.0000        254.0000        255.0000
     sigma:         52.7126         44.4485         52.7339

%%time
histogram(img_cv8, bins=2**8, interval=[0, 2**8 - 1])

Wall time: 1.53 s

print('\n16 bits')
summary(img_cv16)

16 bits
                          R               G               B
       min:        863.0000        863.0000        863.0000
  1st Quar:      23796.0000      23648.0000      23828.0000
    median:      30144.0000      29844.0000      30211.0000
      mean:      33284.9283      31432.4724      33313.5088
  3rd Quar:      42053.0000      36328.0000      42116.0000
       max:      65154.0000      65097.0000      65493.0000
     sigma:      13546.7241      11422.8350      13552.2067

%%time
histogram(img_cv16, bins=2**8, interval=[0, 2**16 - 1])

Wall time: 1.5 s

print('\n32 bits')
summary(img_cv32)

32 bits
                          R               G               B
       min:          0.0010          0.0010          0.0010
  1st Quar:          0.1085          0.1071          0.1088
    median:          0.1788          0.1750          0.1797
      mean:          0.2729          0.2313          0.2734
  3rd Quar:          0.3694          0.2678          0.3706
       max:          0.9868          0.9849          0.9985
     sigma:          0.2392          0.1844          0.2404

%%time
# colorspace linear
histogram(img_cv32, bins=2**8, interval=[0, 1])

Wall time: 1.42 s

%%time
# colorspace sRGB
histogram(img_cv32**(1/2.2), bins=2**8, interval=[0, 1])

Wall time: 1.55 s

#cv2.imwrite('_output/image_8bit_cv.tiff', img_cv8)
#cv2.imwrite('_output/image_16bit_cv.tiff', img_cv16)
#cv2.imwrite('_output/image_32bit_cv.tiff', img_cv32)

ImageIO

---

Reading and writing images using ImageIO.

Installation

---

Install by command line:

>>> conda install -c anaconda imageio
>>> conda install -c dlr-sc freeimageplus

or

>>> pip install imageio
>>> pip install smc.freeimage

Input

---

import imageio as iio

%%time
try:
    img_iio = iio.imread('../_data/woman03.png')
except:
    print('Can not load image!')

Wall time: 16 ms

Analysis

---

summary(img_iio)

                          R               G               B
       min:          0.0000          0.0000          0.0000
  1st Quar:         45.0000         20.0000         16.0000
    median:         64.0000         42.0000         37.0000
      mean:         75.8258         52.5630         46.3073
  3rd Quar:        104.0000         77.0000         67.0000
       max:        255.0000        236.0000        224.0000
     sigma:         43.3170         42.1315         38.3552

%%time
histogram(img_iio)

Wall time: 1.36 s

Output

---

#iio.imwrite('_output/image_iio.tiff', img_iio)

Images 16 e 32 bits

---

%%time
try:
    img_iio8 = iio.imread('../_data/sphere_8bits.tiff')
except:
    print('Can not load 8 bits image!')

try:
    img_iio16 = iio.imread('../_data/sphere_16bits.tiff')
except:
    print('Can not load 16 bits image!')

try:
    img_iio32 = iio.imread('../_data/sphere_32bits.tiff')
except:
    print('Can not load 32 bits image!')

Wall time: 45 ms

print('8 bits')
summary(img_iio8)

8 bits
                          R               G               B
       min:          3.0000          4.0000          3.0000
  1st Quar:         93.0000         92.0000         93.0000
    median:        117.0000        116.0000        118.0000
      mean:        129.5133        122.3053        129.6245
  3rd Quar:        164.0000        141.0000        164.0000
       max:        254.0000        254.0000        255.0000
     sigma:         52.7126         44.4485         52.7339

%%time
histogram(img_iio8, bins=2**8, interval=[0, 2**8 - 1])

Wall time: 1.49 s

print('\n16 bits')
summary(img_iio16)

16 bits
                          R               G               B
       min:        863.0000        863.0000        863.0000
  1st Quar:      23796.0000      23648.0000      23828.0000
    median:      30144.0000      29844.0000      30211.0000
      mean:      33284.9283      31432.4724      33313.5088
  3rd Quar:      42053.0000      36328.0000      42116.0000
       max:      65154.0000      65097.0000      65493.0000
     sigma:      13546.7241      11422.8350      13552.2067

%%time
histogram(img_iio16, bins=2**8, interval=[0, 2**16 - 1])

Wall time: 1.52 s

print('\n32 bits')
summary(img_iio32)

32 bits
                          R               G               B
       min:          0.0010          0.0010          0.0010
  1st Quar:          0.1085          0.1071          0.1088
    median:          0.1788          0.1750          0.1797
      mean:          0.2729          0.2313          0.2734
  3rd Quar:          0.3694          0.2678          0.3706
       max:          0.9868          0.9849          0.9985
     sigma:          0.2392          0.1844          0.2404

%%time
# colorspace linear
histogram(img_iio32, bins=2**8, interval=[0, 1])

Wall time: 1.45 s

%%time
# colorspace sRGB
histogram(img_iio32**(1/2.2), bins=2**8, interval=[0, 1])

Wall time: 1.48 s

#iio.imwrite('_output/image_8bit_iio.tiff', img_iio8)
#iio.imwrite('_output/image_16bit_iio.tiff', img_iio16)
#iio.imwrite('_output/image_32bit_iio.tiff', img_iio32)