Basic Image Noise and Filter Tweaks

Image Noise and Filter

Basic Image Manipulations

Adding noise, applying filters

In [19]:
# Load packages
import os
from PIL import Image, ImageFilter
import numpy as np
import matplotlib.pyplot as plt
from scipy.ndimage import filters

# Ensure plots embeded in notebook
%matplotlib inline
In [2]:
# Open original image
img = Image.open('ducks.jpg','r')
# Cannot use Image show directly
#img.show() # Image will show in external window
# Use numpy array, convert to gray scaling
im = np.array(img.convert('L'))
# Matplotlib doesn't support bmp natively
# Need to change color map to grayscale and set vmin, vmax to reproduce the same luminace of original
plt.imshow(im, cmap='gray',vmin=0, vmax=255)
plt.show()
In [5]:
# Add Noise (Gaussian)
noise_sigma = 30
noise = noise_sigma * np.random.randn(*im.shape)
# the histogram of the noise
plt.subplot(1,2,1)
n, bins, patches = plt.hist(noise,20)
# show the ducks with noise
noisy_ducks = im + noise;
plt.subplot(1,2,2)
plt.imshow(noisy_ducks, cmap='gray', vmin=0, vmax=255)
plt.show()
# Left shows noise is a normal distribution
# Right shows the image with noise added
In [23]:
# Apply Gaussian smoothing kernal to noisy image
filter_sigma = 3
smooth_ducks = filters.gaussian_filter(noisy_ducks,filter_sigma)
plt.imshow(smooth_ducks, cmap='gray', vmin=0, vmax=255)
plt.show()
# Noisy image becomes smooth, or blurred
In [24]:
# Sharpen the blurred image 1: Using PIL built in filter
sharp_img = Image.fromarray(np.uint8(smooth_ducks)).filter(ImageFilter.SHARPEN)
plt.imshow(sharp_img, cmap='gray', vmin=0, vmax=255)
plt.show()
In [20]:
# Sharpen the blurred image 2: using custom filter
# Multiplying each piexl by 2, then subtracting average value of neighbors (bluring filter)
# Overall, the impact is a sharpening filter ("unsharp mark", a term come from old days with dark room)
# It accenturates the difference with local average

# This section needs opencv installed, which mine does not right now, skipped for now
In [25]:
# Sharpen the blurred image 3: using custom filter
# Similar to the above one, but directly use gaussian filter twice to create similar effect
alpha = 30
filter_blurred_image = filters.gaussian_filter(smooth_ducks, 1)
sharp_img_3 = smooth_ducks + alpha * (smooth_ducks - filter_blurred_image)
plt.imshow(sharp_img_3, cmap='gray', vmin=0, vmax=255)
plt.show()
In [ ]:
 

Published: September 08 2016

  • category:
  • tags:
blog comments powered by Disqus