Brett
www.Paufler.net

Lightning Talk
Image Cracking with Python

Why Images?

Motivates Learning
Fun - Easy - Accessible

Basic Data Type
MRI - Xray
Microscopes - Telescopes

Signal Processing
Data Compression
(Entropy & Information Theory)
Computer Vision -> openCV

numpy -> Python's version of Fortran
skimage -< Scientific Python Library

Python Just One Tool in the Box

Irfanview
Inkscape

Collection of Watermarks — Watermark Collection

IrfanView - Resize, Add Watermarks
Inkscape - Make Watermarks & Graphics

Run Irfanview as a subprocess to batch resize, etc.
I find it more convenient than PIL
(Python Image Library).

Eclipse with PyDev

The most important tool in the chain.

Eclipse IDE

Eclipse IDE throwing Error

No foo() defined.
Eclipse throwing an error.

Python Image Cracking Code

Fastest way to get to the innards of an image in Python.


from skimage.io import imread, imsave

img = imread('./input/alcatraz.png')
img = img     #conversion code here
imsave('./output/alcatraz.png', img)

Batch Code With Bells and Whistles


from os import listdir
from skimage.io import imread, imsave

for image in listdir('./input'):

    image = './input/' + image
    print image
	
    img = imread(image)
    print img
	
    sN = image.replace('input', 'output')
    print sN
    imsave(sN, img)

image = path to image file
img = numpy img array
sN = save name

Print Output

Output of Print statement

Three stacked arrays.
Range of integers from 0-255.

Round Trip similiar to 'Hello World'.
Proof setup is working.
Scrapes meta-data.

Is It Working?

Dissecting the Numpy img array.


from skimage.io import imread, imsave

#NOTE: img is a numpy array
img = imread('./input/alcatraz.png')

#Numpy Slicing Syntax
r = img[:,:,0]
g = img[:,:,1]
b = img[:,:,2]

imsave('./output/red_as_grey.png', r)
imsave('./output/green_as_grey.png', g)
imsave('./output/blue_as_grey.png', b)

Grayscale Color Layers

Greyscale of red layer of alcatraz — Greyscale Red

Greyscale of green layer of alcatraz — Greyscale Green

Greyscale of blue layer of alcatraz — Greyscale Blue

Red as Red, etc.


from skimage.io import imread, imsave
import numpy as np

#NOTE: img is a numpy array
img = imread('./input/alcatraz.png')

#Numpy Slicing Syntax
r = img[:,:,0]
g = img[:,:,1]
b = img[:,:,2]

#Create a matching array of zeroes
z = np.zeros_like(r)

#Reconstructing a three layer (R,G,B) img
red = np.dstack((r, z, z))
green = np.dstack((z, g, z))
blue = np.dstack((z, z, b))

imsave('./output/red_layer.png', red)
imsave('./output/green_layer.png', green)
imsave('./output/blue_layer.png', blue)

print r == g
print img.shape

Color Layers

Output of:
print r == g
print img.shape

Numpy r==g output

And Now for Something Completely Different

In which the power of Numpy
is more completely demonstrated.
And hopefully something cool is outputted.


from skimage.io import imread, imsave
import numpy as np

img = imread('./input/alcatraz.png', as_grey=True) 
print 'Grayscale'
print img

#Equal to * 1 below
img = np.array(255 * img, dtype=int)

img_list = []
for n in [1, 2, 3, 5]:
    img = img * n
    print '\n\nimg * %d' % n
    print img
    #create the color img array
    z = np.zeros_like(img)
    g = np.dstack((z, img, z))
    img_list.append(img)
    imsave('./output/green_squared_%d.png' % n, g)

coolness = np.dstack(img_list[1:])
imsave('./output/coolness.png', coolness)

Bug/Error in Above
As I should have assigned the multiplied image to a new variable.

Grayscale printed feedback

grayscale output

Filter Effect Output

Green Effect img *1 — img * 1
* 1 in Total

Green Effect img *2 — img * 2
* 2 in Total

Green Effect img *3 — img * 3
* 6 in Total

Green Effect img *5 — img * 5
* 30 in Total

Tying It All Together

Coolness!

Refactored Code

I may be wordy,
but code shouldn't be...



from skimage.io import imread, imsave
import numpy as np

def image_crack_demo_grey(image='./input/alcatraz.png',
                     sN='./output/alcatraz.png',
                     factors=(2,3,5)):
    '''Grayscale 'coolness' effect applied to passed image file.
        factors are amount (r,g,b) color layers are multiplied by'''
    img = np.array(255 * imread(image, as_grey=True), dtype=int) 
    r = img * factors[0]
    g = img * factors[1]
    b = img * factors[2]
    imsave(sN, np.dstack((r,g,b)))

def image_crack_demo(image='./input/alcatraz.png',
                     sN='./output/alcatraz.png',
                     factors=(2,3,5)):
    '''Color Layer 'coolness' effect applied to passed image file.
        factors are amount (r,g,b) color layers are multiplied by'''
    img = imread(image)
    img[:,:,0] *= factors[0]
    img[:,:,1] *= factors[1]
    img[:,:,2] *= factors[2]
    imsave(sN, img)

def run_image_demo():
    '''Demo run of grayscale and color image_crack_demo
    used to populate grand finale of webpage.'''
    for factors in [(2,3,5), (5,3,2), (2,6,30), (30,6,2)]:
        grey_sN = './output/alcatraz_%d_%d_%d_grey.png' % factors
        image_crack_demo_grey(sN=grey_sN, factors=factors)
        color_sN = './output/alcatraz_%d_%d_%d_color.png' % factors
        image_crack_demo(sN=color_sN, factors=factors)

run_image_demo()

One Liners

After bulk imports:


from skimage.io import imread, imsave
from skimage.filter.rank.generic import entropy
from skimage.filter import sobel
from skimage.morphology import disk

Replacing one liners for img = img

Entropy Filter Example — img = entropy(img, disk(10))

THreshold Filter Example — img[img<=0.64] = 0

Final Thoughts?
How About a Sound File...

Images are signals.
And so is most everything else...

Cracking a sound file is just as easy.


from scipy.signal import resample
from scipy.io.wavfile import read, write

r, d = read(wav_file_in)

Brett www.Paufler.net Lightning Talk Image Cracking with Python