Common / refined code for this chapter

Imports



In [7]:

    
from __future__ import division
%pylab inline
from scipy import ndimage
img = imread('MP.tif')[...,0]









    



Populating the interactive namespace from numpy and matplotlib

Widgets

Use with:

def f(var1, var2, ...):
    # do something with variables

interact(f, var1=(start,stop,step), var2=True, var3=3, ...)

See using interact for more.



In [17]:

    
from IPython.html.widgets import interact

Implemented algoritm and often used functions

Show image shorthand



In [2]:

    
def gimshow(img, size=(12,12), title=''):
    "Shows grayscale image."
    figure(figsize=size)
    plt.title(title)
    imshow(img, cmap='gray')

Show histogram shorthand



In [1]:

    
def histshow(img):
    "Shows a histogram figure of image."
    hist = np.bincount(img.flatten())
    figure()
    title('Levels in image')
    xlim((0, hist.size))
    plot(hist)
showhist = histshow

Histogram equalization

Histogram equalization algoritm, for improving detail in image.



In [4]:

    
def equalization(img):
    "Return histogram equaliated image."
    hist = np.bincount(img.flatten())
    cdf = (hist / img.size).cumsum()
    equal_img = np.zeros(img.shape, dtype=np.uint8)
    for level in range(1,256): # exclude zeroes
        if hist[level] == 0: continue # short cut
        mask = (img == level)
        equal_img[mask] = cdf[level] * 255
    return equal_img

Smoothing

Function which smooth image with convolution.

Note about impletation

Some fractions are impossible to store accurate in binary (and also other number systems). Here is one example:

$1/9 = (0.1111...)_{10}$ $ = (0.000111000111...)_2$ $ = (0.1)_3$

Which is the normalization used for 3x3 mask. Since np.convolute is linear, a rounding error will prevail. Therefor we add 1e-12 to all cells accomodate for this, assuming steps (normalize constant) are far above 1e-12.



In [6]:

    
def smooth(img, size=3):
    """
    Smooths image by convolution. Default mask is 3x3.
    
    Variables
    ---------
    img : ndarray
        Input image to smooth.
    size : int
        Size of square smoothing mask. Should be 2n+1.
        
    Returns
    -------
    result : ndarray
        Smoothed image.
    """
    mask = np.ones((size,size))
    mask /= mask.sum() #normalize
    return ndimage.convolve(img+1e-12, mask).astype(np.uint8)

Scale image

As defined in 2.6-10 and 2.6-11 in GW page 79.



In [6]:

    
def scale(img):
    "Scales intensities in image to span whole image."
    s = np.copy(img).astype(np.float)
    s -= s.min()
    s *= 255 / s.max()
    return s.astype(np.uint8)