Template matching using Normalized Cross Correlation

This program demonstrate the implementation of conventional cross correlation and normalized cross correlation metric to find the similarity score between template and the image portion.This program can be used for image registration to align the given images according to correlated pixels. Also this image demonstrate the use of correlation techniques to match the object. This method is very crude that it may fail most of the time if the object selected to find is a bright image.

Contents

• Start
• Read the image
• Read the Template
• Finding correlation map
• Finding Normalized Cross correlation map
• Display the object in image

Start

clear all;
close all;
clc;

Read the image

image = imread('dollar.tif');
x = double(image)/255;
imshow(x); pause(1);

Read the Template

template = imread('temp.tif');
temp = double(rgb2gray(template))/255;
imshow(temp); pause(1);

Finding correlation map

Cxt = conv2(x,rot90(conj(temp)),'same'); % Cross correlation
imshow(Cxt/max(max(Cxt))); pause(1)

Finding Normalised Cross correlation map

h = ones(size(temp));
Cgg = conv2(x,h,'same'); % finding the energy of signal
NCC = Cxt./Cgg;
imshow(NCC/max(max(NCC))); pause(0.5);

Display the object in image

Locate the maxima [m,n] from the figure and then display the box around the object

[a b] = max(max(NCC));
[c d] = max(NCC);
m = d(b);
n = b;
clear a b c d;
[a b] = size(x);
[c d] = size(temp);
c = round(c/2);
for i=1:a
    for j=1:b
        if (i==m-c && j>n-c && j<n+c)||(i==m+c && j>n-c && j<n+c)...
                || (j==n-c && i>m-c && i<m+c)||(j==n+c && i>m-c && i<m+c)
            x(i,j) = 0;
            x(i+1,j+1) = 0;
        end
    end
end
imshow(x);