Some Notes for Image Processing (2)

Identity Transformation

The output value of intensity is the same with input intensity. So there is no change on the image.

s = r

imageprocess2_01

 

Image Negative

The output value of intensity = (L-1) - input value.

So the dark pixel will be bright pixel, and bright will be dark pixel.

s = (L-1) - r L = 256

imageprocess2_02

Matlab Code

clear all; close all;

f = imread('imageprocess2_03.png');

f_neg = 255 - f; % negative

figure, imshow(f,[]);

figure, imshow(f_neg,[]);

Before

imageprocess2_03

After

imageprocess2_04

 

Contrast Stretching

Producing an image with higher contrast by darkening the intensity levels below k and brightening the levels above k

Matlab Code

clear all; close all;

f = imread('contrast_stretch.png');

f1 = imadjust(f,stretchlim(f),[]);

f2 = imadjust(f, stretchlim(f,[0.05,0.95]),[]);

f3 = imadjust(f, stretchlim(f,[0.10,0.80]),[]);

f4 = imadjust(f, stretchlim(f,[0.20,0.70]),[]);

f5 = imadjust(f, stretchlim(f,[0.49,0.50]),[]);

subplot(2,3,1), imshow(f),title('original image');

subplot(2,3,1), imshow(f1),title('default parameter: 0.01,0.99');

subplot(2,3,1), imshow(f),title('original image');

subplot(2,3,2), imshow(f1),title('default parameter: 0.01,0.99');

subplot(2,3,3), imshow(f2),title('parameter: 0.05,0.95');

subplot(2,3,4), imshow(f3),title('parameter: 0.10,0.80');

subplot(2,3,5), imshow(f4),title('parameter: 0.20,0.70');

subplot(2,3,6), imshow(f5),title('parameter: 0.49,0.50');
imageprocess2_05

 

Power-law (Gamma) transformation

Each input value is raised to the power gamma

imageprocess2_06
  • r < 1 will brighten the image
  • r > 1 will darken the image
imageprocess2_07

Matlab Code

clear all; close all;

f = imread('babytest.png');

f1 = im2double(f);

f1 = power(f1,0.9);

f1 = mat2gray(f1);

f2 = power(f1,0.8);

f2 = mat2gray(f2);

f3 = power(f1,0.7);

f3 = mat2gray(f3);

f4 = power(f1,0.6);

f4 = mat2gray(f4);

f5 = power(f1,0.5);

f5 = mat2gray(f5);

subplot(2,3,1), imshow(f),title('original image');

subplot(2,3,2), imshow(f1),title('parameter: 0.9');

subplot(2,3,3), imshow(f2),title('parameter: 0.8');

subplot(2,3,4), imshow(f3),title('parameter: 0.7');

subplot(2,3,5), imshow(f4),title('parameter: 0.6');

subplot(2,3,6), imshow(f5),title('parameter: 0.5');
imageprocess2_08

 

Bit-plane Slicing

Pixel intensity is represented by bits, and a 256-level grayscale image is represented by 8 bits.

  • The higher-order bit planes (e.g., 8-6) contain the majority significant data.
  • The lower-order bit planes (e.g., 3-1) contribute to more subtle intensity details.

Matlab Code

clear all; close all;

f = imread('bg_old.jpg');

clear all; close all;

image = imread('bg_old.jpg'); % convert colourful image to gray image

f = im2gray(image);

% gets bit plane 

f1 = bitget(f,1); 

f2 = bitget(f,2); 

f3 = bitget(f,3); 

f4 = bitget(f,4); 

f5 = bitget(f,5); 

f6 = bitget(f,6); 

f7 = bitget(f,7); 

f8 = bitget(f,8); 

subplot(3,3,1),imshow(f),title('original image');

subplot(3,3,2),imshow(logical(f1)),title('bit plane 1');

subplot(3,3,3),imshow(logical(f2)),title('bit plane 2');

subplot(3,3,4),imshow(logical(f3)),title('bit plane 3');

subplot(3,3,5),imshow(logical(f4)),title('bit plane 4');

subplot(3,3,6),imshow(logical(f5)),title('bit plane 5');

subplot(3,3,7),imshow(logical(f6)),title('bit plane 6');

subplot(3,3,8),imshow(logical(f7)),title('bit plane 7');

subplot(3,3,9),imshow(logical(f8)),title('bit plane 8');
imageprocess2_09

From the result, we can see that if only use bit plane 1, it's hard to recognize the image. However, with only bit plane 8, it can recognize something.

 

Let's try another code. Let's compare the following conditions:

  • Only using bit plane 8
  • Using bit plane 8 and 7
  • Using bit plane 8, 7, and 6

Matlab Code


clear all; close all;

image = imread('bg_old.jpg');

f = im2gray(image); % convert colourful image to gray image

f1 = zeros(size(f)); % create blank image with the same size of original image

f2 = zeros(size(f));

f3 = zeros(size(f));

% set bit, f1 only uses bit plane 8, f2 uses bit plane 8,7, f3 uses bit plane 8,7,6

f1 = bitset(f1,8,bitget(f,8));

f2 = bitset(f2,8,bitget(f,8));

f2 = bitset(f2,7,bitget(f,7));

f3 = bitset(f3,8,bitget(f,8));

f3 = bitset(f3,7,bitget(f,7));

f3 = bitset(f3,6,bitget(f,6));

f1 = uint8(f1); f2 = uint8(f2); f3 = uint8(f3); 

subplot(2,2,1),imshow(f),title('original image');

subplot(2,2,2),imshow(f1),title('using bit plane 8');

subplot(2,2,3),imshow(f2),title('using bit plane 8, and 7');

subplot(2,2,4),imshow(f3),title('using bit plane 8, 7, and 6');
imageprocess2_10

 

It shows that with more bit planes, the image is more clear and close to the original image.

 

Histogram Equalization

  • A method for modifying (enhancing) the dynamic range and the conteast of an image

Matlab Code

clear all; close all;

f = imread('satellite.png');

f1 = imadjust(f, stretchlim(f),[]);

f2=histeq(f);

subplot(2,3,1),imshow(f);

subplot(2,3,1),imshow(f), title('original image');

subplot(2,3,2),imshow(f1), title('contrast stretching');

subplot(2,3,3),imshow(f2), title('histogram equalization');

subplot(2,3,4),imhist(f), title('original histogram');

subplot(2,3,5),imhist(f1), title('contrast histogram');

subplot(2,3,6),imhist(f2),title('hist-eq histogram');

 

imageprocess2_11