Manipulating Images in Python



In [ ]:

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

The Python Image Library (PIL)



In [ ]:

    
from PIL import Image

You can load images as `Image` instances



In [ ]:

    
img = Image.open("img/colonies.jpg")
plt.imshow(img)

`Image` instances can be interconverted with `numpy` arrays



In [ ]:

    
arr = np.array(img)
print("x,y,RGB ->",arr.shape)

Values are represented for different channels in bits

8-bit channels have $2^8 = 256$ possible values.
16-bit channels have $2^{16} = 65,6536$ possible values

Major formats you'll encounter

8-bit grayscale (1 channel, 0-255).
8-bit color (3 channels, each 0-255)
8-bit RGBA (4 channels, each 0-255)
16-bit grayscale (1 channel, 0-65535)
16-bit color (3 channels, each 0-65535)
16-bit RGBA (4 channels, each 0-65535)

By convention, channels are arranged: Red, Green, Blue, Transparency

What kind of image did we just read in?

What does the following code do?



In [ ]:

    
arr = np.array(img)
plt.imshow(arr)
plt.show()
arr[:,:,1] = 255

plt.imshow(arr)

Can convert arrays back to images and save



In [ ]:

    
x = Image.fromarray(arr)
x.save('junk.png')

One powerful method is to create a mask



In [ ]:

    
arr = np.array(img)
low_red = arr[:,:,0] < 50 

arr[:,:,0] = low_red*255
arr[:,:,1] = low_red*255
arr[:,:,2] = low_red*255


plt.show()
plt.imshow(arr)

You can combine masks with "*" (and) and "+" (or)

Predict the output of the next cell.



In [ ]:

    
print("True and True =", bool(True*True))
print("False and True =", bool(False*True))
print("Flase and False =",bool(False*False))

print("True or False =",bool(True + False))
print("True or True =",bool(True + True))
print("False or False =",bool(False + False))

Combined mask

What will the following code do?



In [ ]:

    
arr = np.array(img)

red = arr[:,:,0] < 2
green = arr[:,:,1] > 5
blue = arr[:,:,2] < 5
mask = red*green*blue

new_arr = np.zeros(arr.shape,dtype=arr.dtype)
new_arr[:,:,1] = mask*255

plt.imshow(img)

Find pixels with some green but little red or blue, then set those to green. All other pixels are black.

Write a block of code that reads in the image, sets any pixel with red > 20 to (0,0,0), and then writes out a new .png image.



In [ ]:

    
img = Image.open("img/colonies.jpg")
arr = np.array(img)
mask = arr[:,:,0] > 20

arr[mask,0] = 0
arr[mask,1] = 0
arr[mask,2] = 0

plt.imshow(arr)
img2 = Image.fromarray(arr)
img2.save("junk.png")



In [ ]: