Digital Image Processing

& Pattern Recognition

Lecture#3
Image Smoothing & Noise Removal

Prepared by: Dr. Gawed Nagi

Contents
This lecture will cover:
◼ Histogram Matching (Specification)
◼ Image Smoothing Filters
 Mean Filter
 Median Filter

 Gaussian Filter

◼ Point Processing
 Image negatives
 Log Transformation

 Power-Law (Gamma) Transformation

 Piecewise Linear Transformation Functions

Histogram Matching (Specification)
 Assume we have two images and each has
its specific histogram. So
Is it possible to modify one image based on
the contrast of another one? YES
 In fact, this is the definition of the
histogram matching.
 In other words, given images A, and B, it
is possible to modify the contrast level of
A according to B.
Histogram Matching (Specification)
 In order to match the histogram of images
A and B, first equalizing the histogram of
both images. Then,
 Mapping each pixel of A to B using the
equalized histograms. Then
 Modifing each pixel of A based on B.
Histogram Matching (Specification)

Original Image Target Image Matched Image

Histogram Matching (Specification)
MATLAB Hints
% Example of Histogram Matching
A = imread("in.jpg");
Ref = imread("ref.jpg");
B = imhistmatch(A,Ref);

montage({A,Ref,B})
title('Input Image (Left) vs Reference Image (Right) &
Output Image (Down)’);
%{
subplot(2, 2, 1), imshow(A), title ('Input Image');
subplot(2, 2, 2), imshow(Ref), title ('Reference Image');
subplot(2, 2, 3), imshow(B), title ('Output Image');
%}
Image Enhancement Domains
Image Enhancement Domains

Neighborhood
Histogram Point
Processing
Processing Processing
(Filtering)
Image Smoothing & Enhancing
 Image smoothing, Noise removing, edge
enhancement and boundary detection are
local operations in image processing.
 A Local operation is an operation that
transfer input image into an output image
depending on the neighborhood values of
input coordinates.
 It means, if pixel b[x,y] is a pixel in
output image, it will be result based on
the a[x,y], the same pixel in input image
and the neighborhood pixel values.
Neighbors of A Pixel

❖some of the points in ND (p) and N8 (p) fall outside the

image if (x,y) is on the border of the image.
Smoothing Spatial Filtering
Spatial
Filtering

Non-Linear
Linear Filtering
Filtering

Mean/Averaging
Median Filter
Filter

Max Filter

Gaussian Filter Min Filter

Linear Filtering
Convolution
convolution is simply the process of taking
a small matrix called the kernel and
running it over all the pixels in an image.
 At every pixel, we’ll perform some math
operations involving the values in the
convolution matrix and the values of a
pixel and its surroundings to determine
the value for a pixel in the output image.
Mean / Averaging Filtering
(Neighborhood Averaging)
 When an image is acquired by a camera or
other imaging system, normally the vision
system for which it is intended is unable to
use it directly.
 The image may be corrupted by random
variations in intensity, variations in
illumination, poor contrast or noise that
must be handle with in the early stages of
vision processing.
 Mean / Averaging Filtering
 Therefore, mean filter is one of the
techniques which is used to reduce noise of
the images.
 This is a local averaging operation and it is a
one of the simplest linear filter.
 The value of each pixel is replaced by the
average of all the values in the local
neighborhood. Let f(i,j) is a noisy image
then the smoothed image g(x,y) can be
obtained by,

Where S is a neighborhood of (x,y) and n is the number of pixels in S.

Mean / Averaging Filtering
 3x3 mean filtering,

1/9
Mean / Averaging Filtering
Mean / Averaging Filtering
 Kernel 5x5
MATLAB Hints
% mean filter with 3X3 mask/window size
I = imread("Lena2.png");
N = imnoise(I,'salt & pepper',0.03);
K = ones(3,3)/9;
noise_free = imfilter(N,K);
subplot(2,2,1), imshow(I), title ('Original Image');
subplot(2,2,2), imshow(N), title ('Noisy Image');
subplot(2,2,3), imshow(noise_free), title ('After
Noisy Removal');
Median Filter
 Median, in statistics, is the middle value of
the given list of data when arranged in an
order.

X = ordered list of values in data set

N = number of values in data set
Median Filter
Median Filter
MATLAB Hints
I = imread ("Lena2.png");
N = imnoise(I,'salt & pepper', 0.3);
red_channel = N (:, :, 1);
green_channel = N (:, :, 2);
blue_channel = N (:, :, 3);
red_channel = medfilt2(red_channel , [3 3]);
green_channel = medfilt2(green_channel , [3 3]);
blue_channel = medfilt2(blue_channel , [3 3]);
F = cat (3, red_channel , green_channel ,
blue_channel );
subplot(2, 1, 1), imshow(N), title ('Noisy Image');
subplot(2, 1, 2), imshow(F), title ('Image After Noise
Removal');
Gaussian Filter
 instead of a box filter consisting of equal
filter coefficients, a Gaussian kernel is used.
 Gaussian filtering is highly effective in
removing Gaussian noise from the image.
 Gaussian filtering is used to blur images and
remove noise and detail.
Gaussian Filter
 Gaussian function in two dimensions:

σ – the standard deviation

Gaussian Filter
Source Pixels
New Pixels
Gaussian Filter
 5x5 Gaussian kernel
 Finding out the values of a Gaussian Kernel
 If the kernel size is 5x5 and the value of σ =1, We
have to find out all the values for this kernel. The
equation of Gaussian function is:

 For a filter size x by y, the number of rows are x and

the number of columns are y.
 Generally, The filter size consists of odd numbers, In
this case ,the values for rows are
-{(x-1)/2} to +{(x-1)/2}
and the values of columns are
-{(y-1)/2} to +{(y-1)/2} .
 Finding out the values of a Gaussian Kernel
 So, for filter size 5x5 , the row values are -2 to +2
and column values are -2 to +2. So, if we want to
calculate the 1st value of the kernel , we have to put
the row value and column value of the kernel for
that position in the Gaussian Function.
.
Finding out the values of a Gaussian Kernel
 we can see that all the values are very small, if
we normalize all the values by dividing all the
values by 0.003 , we will get the following table:
MATLAB Hints
I = imread ("Lena2.png");
N = imnoise(I,'salt & pepper', 0.05);

red_channel = N (:, :, 1);

green_channel = N (:, :, 2);
blue_channel = N (:, :, 3);

g_Filter = fspecial (‘gaussian’, [5 5], 4);

red_channel = imfilter(red_channel , g_Filter );

green_channel = imfilter(green_channel , g_Filter );
blue_channel = imfilter(blue_channel , g_Filter );
MATLAB Hints
F = cat (3, red_channel , green_channel , blue_channel );

subplot(2, 1, 1), imshow(N), title ('Noisy Image');

subplot(2, 1, 2), imshow(F), title ('Image After Noise
Removal by Gaussian Filter');
Point Processing

Neighborhood
Histogram Point
Processing
Processing Processing
(Filtering)
Point Processing
Basic Gray Level Transformations:
 image enhancement can be done by gray-
level transformation functions.
 These are among the simplest of all image
enhancement techniques.
 The values of pixels, before and after
processing, will be denoted by r and s,
respectively. s=T(r) Or
Where r is the pixels of the input image and s is the pixels of the
output image. T is a transformation function that maps each
value of r to each value of s.
Intensity Transformation Functions
Image Negatives /The Negative
Transformation
 The negative of an image with gray levels
in the range [0, L-1] is obtained by using
the negative transformation
s = L - 1 - r.
 The negative of an image with gray levels
in the range [0, L-1] is obtained by using
the negative transformation
Image Negatives /The Negative
Transformation
Image Negatives /The Negative
Transformation
The Negative Transformation
The Negative Transformation
MATLAB Hints
% reading the RGB file into the Matlab environment
I = imread("lena2.png");
subplot(1, 2, 1),
% displaying the RGB image
imshow(I);
title("Original image");
% levels of the 8-bit image
L = 2 ^ 8;
% finding the negative
neg = (L - 1) - I;
subplot(1, 2, 2),
% displaying the negative image
imshow(neg);
title("Negative Image")
Log Transformation
 Log transformation to visualize patterns in
the dark regions of an image.
 When the input grey level values may
have an extremely large rang of values.
 E.g. Fourier Transform can have values in
the rang [0-106]
Log Transformation

Log transform inverse

Log Transformation
 In the log transformation, the low-intensity
values are mapped into higher intensity
values.
 It maps a narrow range of low gray levels to
a much wider range.
 The inverse log transform is opposite to log
transform. It maps a narrow range of high
gray levels to a much wider range.
 The inverse log transform expands the
values of light-level pixels while
compressing the darker-level value
Log Transformation
Log Transformation
 MATLAB Hints
% MATLAB program to demonstrate log transformation of image
input_image =imread(‘tst.jpg’);
% Convert the image to double datatype for calculations
input_image = im2double(input_image);
% Constant to determine the nature of the log curve
c = 1;
% Perform the log transformation
log_transformed = c * log(1 + input_image);
% Display the original image and log-transformed image
subplot(1, 2, 1), imshow(input_image), title('Original Image’);
subplot(1, 2, 2), imshow(log_transformed), title('Log-
Transformed Image’);
% Save the log-transformed image
imwrite(log_transformed,'log_transformed_image.jpg');
Power-Law (Gamma) Transformations
 Power-law transformations have the form

Variation in the value of γ varies the enhancement

of the images. Different display devices / monitors
have their own gamma correction, that’s why they
display their image at different intensity.
Power-Law (Gamma) Transformations
Power-Law (Gamma) Transformations
Power-Law (Gamma) Transformations
Power-Law (Gamma) Transformations
Power-Law (Gamma) Transformations
Power-Law (Gamma) Transformations
 MATLAB Hints
% MATLAB program to demonstrate log transformation of image
input_image =imread(‘tst.jpg’);
% Convert the image to double datatype for calculations
input_image = im2double(input_image);
% Set a desired gamma value for the power law transformation
gamma = 0.4;
% Perform the power law transformation of image
power_law_image = input_image .^ gamma;
% Display the original and power law transformed images
subplot(1, 2, 1), imshow(input_image), title('Original Image’);
subplot(1, 2, 2), imshow(power_law_image), title('Power Law
Transformed Image’);
% Save the power law transformed image
imwrite(power_law_image,'power_law_transformed_image.jpg');
Piecewise Linear Transformation
Functions
Q&A