Image-smoothing techniques-Part1 can be found here

Median Blurring:

In this technique, median of the kernel space is used to replace the value of every pixel( which is the center of that kernel ). As we perform operation on the center of the kernel space hence the width and height of the space is always positive odd number.  

In median blurring, the central pixel is always replaced by some pixel value which is present in the image. This property is not present in other techniques like Averaging and Gaussian blurring.

Median blurring supports inplace operation and is effective in smoothing salt and pepper noise.

cv2.medianBlur( source, ksize ) function in OpenCV is used to smooth the image using median blurring. The function has two arguments:

  1. source: it signifies the source image on which smoothing has to be done.
  2. ksize: it signifies the size of kernel matrix. The value of ksize must be odd and greater than 1 always.

Implemented code in Python:

# Importing cv2 
import cv2 

#read input image 
img = cv2.imread('cat.jpg');

#increasing ksize will make the image smoother
blur1 = cv2.medianBlur(img,5); 
blur2 = cv2.medianBlur(img,7);

cv2.imshow('Original',img);
cv2.imshow('blurred image 1',blur1);
cv2.imshow('blurred image 2',blur2);

#wait for 10 sec
cv2.waitKey(10000);
cv2.destroyAllWindows();

OUTPUT:

Bilateral Filtering:

As we have observed in all techniques of smoothing that edges and noises are smoothed using some function. But the main purpose of smoothing is to remove noise and not to make the whole image blurred. Hence we need some technique that won’t make the edges blur.

Bilateral filtering ensures to not to affect the edges while smoothing the image by considering the intensities of neighboring pixels. It also uses Gaussian function to prepare kernel matrix but it considers only that neighboring pixels in the kernel which have similar intensity to the central pixel. More precisely bilateral filtering is based on Gaussian function with intensity difference.

As on edges the neighboring pixels will have a drastic difference of intensity as compared to central pixel, hence the edges are preserved from smoothing.

But bilateral filtering is comparatively slow than other filters and it does not support inplace operation.

cv2.bilateralFilter(source, ksize, sigmaColor, sigmaSpace) function in openCV is used to perform bilateral filtering. The function has four arguments:

  1. source: it signifies the source image on which smoothing has to be done.
  2. ksize:  it signifies the size of kernel matrix. The value of ksize must be odd and greater than 1 always. When ksize<=0 ksize is proportional to sigmaSpace.
  3. sigmaColor: it signifies the filter in color space. A larger value of the parameter means that farther colors within the pixel neighborhood will be mixed together resulting in larger areas of semi-equal color.
  4. sigmaSpace: it signifies the filter in coordinate space. A larger value of the parameter means that farther pixels will influence each other as long as their colors are close enough.

Implemented code in Python:

# Importing cv2 
import cv2 

#read input image 
img = cv2.imread('door.jpg');

blur1 = cv2.bilateralFilter(img,5,75,100); #ksize = 5
blur2 = cv2.bilateralFilter(img,-1,100,120); #ksize<=0 hence kernel size is calculated using sigmaSace

cv2.imshow('Original',img);
cv2.imshow('blurred image 1',blur1);
cv2.imshow('blurred image 2',blur2);

#wait for 10 sec
cv2.waitKey(10000);
cv2.destroyAllWindows();

OUTPUT:

2D Convolution:

It is actually an image filtering technique which can also be used to remove noise and make it smooth. An arbitrary low pass linear filter is applied on the image to create a blurring effect. 

cv2.filter2D( source, depth, kernel ) function in openCV is used to perform filtering on images. The function has three arguments:

  1. source: it signifies the source image on which smoothing has to be done.
  2. depth:  desired depth of the output image.
  3. Kernel: it specifies the kernel filter matrix.

Implemented code in Python:

# Importing libraries
import cv2 
import numpy as np
 
#read input image 
img = cv2.imread('lady.jpg');
 
kernel = np.ones((4,4),np.float32)/16;
 
blur = cv2.filter2D(img,-1,kernel);
 
cv2.imshow('Original',img);
cv2.imshow('blurred image',blur);
 
#wait for 10 sec
cv2.waitKey(10000);
cv2.destroyAllWindows();

Thank you Rishika Gupta for this article.


0 Comments

Leave a Reply

Your email address will not be published. Required fields are marked *

Insert math as
Block
Inline
Additional settings
Formula color
Text color
#333333
Type math using LaTeX
Preview
\({}\)
Nothing to preview
Insert