How do I show on a figure the the result of histogram intersection of two images?

Hello,

I have selected from a video two frames and I want to use the Histogram Intersection.

When I run the code I get the values [1;256] I'm guessing that the intersection is between those to values.

I would like to know if my code is correct because I'm a bit confuse about the result.

Also I would like to know is there is a way to show the intersection of those to images in a histogram.

Please find below my code

function [output]=calculate_histIntersect(filename,threshold)

filename='C:\Users\User\Videos\videos\TestSet\01_DevilsAdvocate_01.mp4';

vObj= VideoReader(filename);

%From our video read frame 40

% convert the image to grayscale

vFrame1 = read(vObj, 40);

vFrame1grey = rgb2gray(vFrame1);

%From our video read frame 135

% convert the image to grayscale

vFrame2 = read(vObj, 135);

vFrame2grey = rgb2gray(vFrame2);

% Computing the threshold for both frames

% Define threshold value 40

%create a matrix of mrow and ncolum this is done for both brames

%create a matrix for the threshold images

threshold=40;

[mrow, ncol]=size(vFrame1grey);

[urow, vncol]=size(vFrame2grey);

ThvFrame1grey=zeros(mrow, ncol);

ThvFrame2grey=zeros(urow, vncol);

% arrange the value of the first frame according to the threshold

% If the values of the original image are bigger than threshold in the new

% image asign value 1 if not asign value0

for i=1:mrow

for j=1:ncol

if vFrame1grey(i,j)> threshold

ThvFrame1grey (i,j)=1;

else ThvFrame1grey (i,j)=0;

end

end

end

% arrange the value of the second frame according to the threshold

% If the values of the original image are bigger than threshold in the new

% image asign value 1 if not asign value0

for i=1:urow

for j=1:vncol

if vFrame2grey(i,j)> threshold

ThvFrame2grey (i,j)=1;

else ThvFrame2grey (i,j)=0;

end

end

end

% Compute the histogram for the thresholds images

hist01=imhist(ThvFrame1grey);

hist02=imhist(ThvFrame2grey);

% Create absolute value dor both histograms

hist01abs=abs(hist01);

hist02abs=abs(hist02);

% Compute histogram intersection

histIntersect= min(hist01,hist02)/min(hist01abs,hist02abs);

Intersect = find(histIntersect);

output = sort(Intersect);

figure

subplot(2,2,1),imshow(vFrame1grey)

subplot(2,2,2),imhist(vFrame1grey)

subplot(2,2,3),imshow(vFrame2grey)

subplot(2,2,4),imhist(vFrame2grey)

figure

subplot(2,2,1),imshow(ThvFrame1grey)

subplot(2,2,2),imhist(hist01)

subplot(2,2,3),imshow(ThvFrame1grey)

subplot(2,2,4),imhist(hist02)

end

[Image Analyst]

What is the algorithm? Is it this:

histIntersect= min(hist01,hist02)/min(hist01abs,hist02abs);

Who wrote this code? It's fairly well commented but the code is inefficient (where it's right), and very wrong in a few places. I don't want to "fix" that code because I still feel like it would not be right. Do you have something like a Wikipedia page decribing the algorithm?

For examples, your two double nested for loops with if blocks could all be done with 2 lines of code much faster. And histograms are counts, so they are always positive and no need to take their absolute values. And your min() give vectors so you'd need dot slash, not slash which is a matrix divide, but then that would just give 1 for the output since the absolute value of the histogram is the same as the histogram. I don't think you want to do all this just to compute 1. Etc. for several other problems I haven't mentioned.

[Isida Kaloshi]

Hello,

Thank you for your response!

I wrote the code, I'm a beginner with matlab and I guess I have a lot to learn.

Basicly I wanted to compute the threshold of those to frames and then find the histogram intersection.

I read files for intersection on matlab forum but I guess I didin't do it correctly!

[Image Analyst]

First of all, to binarize an image, you can skip the double for loops and do this

binaryImage = grayImage < someThresholdValue;

Next, I still need to know how you're defining histogram threshold, because the formula you're using, looking at the bins where it's the min height of either histogram, seems like it would give something of unknown meaning. I'd like to know what that computation is supposed to mean. It's not going to tell you how alike the two images are, if that's what you're thinking. You could have two images with identical histograms look totally different. For example just take one image and sort it's pixels in order of ascending pixel value and you'll get an image that looks like a ramp and will have a histogram identical to the original image but look nothing like it whatsoever.

[Isida Kaloshi]

Hello,

Thank you for your reply. I did what you advise me and it is much more easier.

binaryImage = vFrame1grey < 40;

Regarding the the histogram intersection.

Bellow is the paper I was reading, I wanted to apply Swain & Ballard histogram Intersection method, but I'm doing something wrong.

https://www.researchgate.net/publication/221315132_A_Comparison_on_Histogram_Based_Image_Matching_Methods

I change the my code but stiil I think I'm doing something wrong

instead of absolute values I've created normalised value of histgram and then the histogram intersection.

% Create normalised value of the histogram

hist01=hist01/sum(hist01);

hist02=hist02/sum(hist02);

% Compute histogram intersection

histIntersect= min(hist01,hist02);

[Prajith Chilummula]

According to the document shared, the code seems fine. Your updated code should work.

hist01=hist01/sum(hist01);

hist02=hist02/sum(hist02);

% Compute histogram intersection

histIntersect= min(hist01,hist02);

[Isida Kaloshi]

Hello,

Thank you very much for your answer