Balanced Histogram Thresholding with MATLAB. Basic Concept & programming

Balanced Histogram Thresholding in MATLAB

Basic Concept

The main idea is to find the point of the image histogram to balance, the unbalanced histogram of the image. In this method, the image is divided into two main parts, background, and foreground. The method finds the point in the histogram where both parts (background & foreground) are balanced.

The Algorithm

• we determine the start and endpoints in the Histogram of the image
• From these two points, we calculate the midpoint to divide the histogram into the left side & the right side
• we calculate the weights or sum of both parts ( left side & right side) of the histogram
• we compare the sums of both parts
• remove the farthest point (weight) from the heavier side or part
• adjust the midpoint
• recalculate the sums of the left & right side
• compare the weights
• we repeat this procedure until the sums or starting and endpoints become equal or the same on both sides.

The process as shown in a GIF from Wikipedia

MATLAB Program

You can download the code from my GitHub repository.

Or just copy-paste the following code.

 %Reading Image & converting data type to double
 img=imread('LovelySpider.jpeg');
 %img(:,:,[2:3])=[]; %uncomment this if image is rgb
 figure
 subplot(2,1,1)
 imshow(img)
 title('Original Image')
 img=double(img);


 I=img(:);           % Calculating Histogram
 hst=zeros(1,256);
 for ii =0:255
     hst(ii+1)=sum(I==ii);
 end

 for ii=1:256        % Calculating Start Point
     if hst(ii)>0
         stpt=ii;
         break
     end
 end
 for ii=256:-1:1     % Calculating End point
     if hst(ii)>0
         endpt=ii;
         break
     end
 end
 mdpnt=round((stpt+endpt)/2);    %mid point
 lsum=sum(hst(stpt:mdpnt));      % sum of left side
 rsum=sum(hst(mdpnt:endpt));     % sum of right side
 while lsum ~= rsum              % iterative process of finding
     if rsum>lsum                % balanced mid point
         endpt=endpt-1;
         if round((stpt+endpt)/2)< mdpnt
             mdpnt=mdpnt+1;
             lsum=sum(hst(stpt:mdpnt));
             rsum=sum(hst(mdpnt:endpt));
             
         end
     else
         stpt=stpt+1;
         if round((stpt+endpt)/2) > mdpnt
             mdpnt=mdpnt-1;
             lsum=sum(hst(stpt:mdpnt));
             rsum=sum(hst(mdpnt:endpt));
             
         end
     end
            
            
 end

 % image processing
 nimg=zeros(size(img));
 rng=size(img);
 for ii=1:rng(1)
     for jj=1:rng(2)
         if img(ii,jj)<=(mdpnt/2)  % threshold point to separate the image background
             nimg(ii,jj)=255;
         else
             nimg(ii,jj)=0;
         end
     end
 end


 subplot(2,1,2)
 imshow(nimg)
 title('Processed Image')

 I=nimg(:);           % Calculating Histogram
 hst2=zeros(1,256);
 for ii =0:255
     hst2(ii+1)=sum(I==ii);
 end
 figure
 subplot(2,1,1)
 stem(hst)
 grid on
 title('original Image Histogram')
 axis([1 256 0 65000])
 subplot(2,1,2)
 stem(hst2)
 title('processed Image Histogram')

 figure
 stem(hst)
 grid on
 %axis([0 172 0 65000])
 hold on
 stem(mdpnt,hst(mdpnt),'red', 'linewidth',2)
 disp('The balanced threshold value of Histogram is :')
 disp(mdpnt)

Figure 2

Useful Links

My LinkedIn Profile

IoT based SpO2 Monitor

GitHub Profile

Youtube Video