CS 111 f21 — Image Data

1 Polls

What will be printed?1 

s = "abc"
s = s[:1] + "MOOO" + s[1:]
print(s)

What will be printed?2 

data = [[3, 1, 6],
        [2, 8, 9],
        [5, 5, 5]]
print(data[:2][-1][::-1][0])

2 2D Data

  • index multiple times to get to individual elements in a nested list
    • data[:2] is [[3, 1, 6], [2, 8, 9]]
    • data[:2][-1] is [2, 8, 9]
    • data[:2][-1][::-1] is [9, 8, 2]
    • data[:2][-1][::-1][0] is 9
  • row-major order
    • the inner lists are the rows, not the columns
  • don't really behave like a grid with rows and columns
    • try to slice a column: data[:][1] (the idea would be all rows, second column, but instead we just get the second row)
    • a module called numpy provides a very useful variation on a list called an array

3 Image data

  • how to represent color?
    • printing on paper: CMYK (cyan, magenta, yellow, black)

CMYK.png

  • digital: RGB (red, green, blue)
    • on screens, tightly packed red, green, and blue elements light with different intensities

rgb-pixels.jpg

  • many different geometries are possible

pixel-geometry.jpg

  • Samsung Galaxy Note 4

samsung-galaxy-note4-pixels.jpg

  • Samsung Galaxy Nexus

samsung-galaxy-nexus-pixels.jpg

  • Apple iPhone 5

iphone-pixels.jpg

3.1 Represent an Image in Python (2x2 example)

rgb-pixels-2x2.png

3D list! 

[[[1.0, 0.0, 0.0], [0.0, 1.0, 0.0]],
 [[0.0, 0.0, 1.0], [1.0, 0.0, 0.0]]]

4 Working With Images

import matplotlib.pyplot as plt
image = plt.imread("owl.png")
  • image data as an array
    • what you see when you print 
      [[[0.5019608  0.5411765  0.40784314 1.        ]
  [0.49803922 0.5254902  0.39215687 1.        ]
  [0.5019608  0.5137255  0.37254903 1.        ]
  ...
  [0.3764706  0.3529412  0.34901962 1.        ]
  [0.3647059  0.35686275 0.35686275 1.        ]
  [0.37254903 0.37254903 0.37254903 1.        ]]

 [[0.49803922 0.52156866 0.40784314 1.        ]
  [0.5058824  0.5254902  0.40784314 1.        ]
  [0.5058824  0.5137255  0.39215687 1.        ]
  ...
  [0.3647059  0.32941177 0.3372549  1.        ]
  [0.38039216 0.35686275 0.37254903 1.        ]
  [0.3882353  0.36862746 0.3882353  1.        ]]

 [[0.4862745  0.4862745  0.4117647  1.        ]
  [0.48235294 0.48235294 0.40392157 1.        ]
  [0.49411765 0.5019608  0.41960785 1.        ]
  ...
  [0.3647059  0.30980393 0.3372549  1.        ]
  [0.40392157 0.34509805 0.39607844 1.        ]
  [0.40392157 0.34901962 0.40784314 1.        ]]

 ...

 [[0.3254902  0.32941177 0.26666668 1.        ]
  [0.3019608  0.30588236 0.24705882 1.        ]
  [0.3137255  0.32156864 0.27058825 1.        ]
  ...
  [0.4392157  0.4        0.29803923 1.        ]
  [0.4627451  0.41960785 0.3372549  1.        ]
  [0.44705883 0.40392157 0.32941177 1.        ]]

 [[0.29411766 0.28235295 0.22352941 1.        ]
  [0.3254902  0.31764707 0.26666668 1.        ]
  [0.31764707 0.31764707 0.2784314  1.        ]
  ...
  [0.43137255 0.39607844 0.30980393 1.        ]
  [0.41960785 0.3882353  0.30980393 1.        ]
  [0.40392157 0.37254903 0.29411766 1.        ]]

 [[0.34509805 0.32156864 0.26666668 1.        ]
  [0.33333334 0.31764707 0.27058825 1.        ]
  [0.2901961  0.2901961  0.25490198 1.        ]
  ...
  [0.41960785 0.3882353  0.30980393 1.        ]
  [0.4117647  0.38431373 0.30588236 1.        ]
  [0.40784314 0.38039216 0.3019608  1.        ]]]
    • 0 to 1 for each of red, green, and blue
    • fourth value is the transparency, or alpha
    • image.shape gives us a tuple of the dimensions, (702, 1024, 4)
      • 702 rows, 1024 columns, 4 values per pixel
    • image.size gives us the total number of entries in the array
      • \(702\times1024\times4 = 2875392\)
    • len, loops the same 
      for row in range(len(image)):
  for column in range(len(image[row])):
    rgba_pixel = image[row][column]
    • unlike lists, numpy arrays are fixed size, and can hold only one kind of data (i.e., a particular array must be all integers or all strings)
  • array indexing
    • row, column
      • can do image[row, column] instead of image[row][column]
    • slicing along multiple dimensions
      • slice out the owl's face: image[200:500, 200:800]

4.1 Fruit Ninja Redux3

  • crop the image by 200 pixels on each side
  • slice out 100-by-100 box in bottom right corner
  • black out every other column (set all the pixels in every other column to [0.0, 0.0, 0.0, 1.0]

Footnotes:

1

aMOOObc

2

9

3 
import matplotlib.pyplot as plt
image = plt.imread("owl.png")
cropped = image[200:image.shape[0] - 200, 200:image.shape[1] - 200]
plt.imshow(cropped)
plt.show()

box = image[image.shape[0] - 100:, image.shape[1] - 100:]
plt.imshow(box)
plt.show()

blacked_out = image.copy()
for row in range(blacked_out.shape[0]):
  for column in range(0, blacked_out.shape[1], 2):
    blacked_out[row, column] = [0.0, 0.0, 0.0, 1.0]
plt.imshow(blacked_out)
plt.show()