Plotting mean of the columns of image on that image

I am struggling to plot the mean of the columns of the gray scale image on the same image. Mean of the columns of an image can be found by using Mean command in Matlab, but i don't have any idea how to plot it in image coordinates. Any idea or help is appreciated! Thank you :)

[Guillaume]

What do you mean by plotting?

- superimpose a line graph of mean values with scale axes?

- add a row of pixels whose grey value is the mean?

- something else?

[Yawar Rehman]

add a row of pixels whose grey value is the mean, which can divide the image in two portions horizontally.

[Yawar Rehman]

actually, i want to make two portions of an image horizontally by taking the mean of the grey values of its columns some thing like this, the mean (red line) may not be a straight line rather it will be a curve i guess

[Yawar Rehman]

Can anyone help me out in this please!!

[Guillaume]

Adding "a row of pixels whose grey value is the mean" is trivial. However, I don't understand "which can divide the image in two portions horizontally". What do you mean by that?

Similarly, with the image you've shown, I don't understand what the red line is supposed to represent. How do you choose the row where the red points go?

[Yawar Rehman]

I think i have been using the wrong term "plotting" to express myself.

mean function of Matlab returns with the mean of the columns of an image. for finding that where does that column's mean value lie, i guess we have to go through the column pixel search. when there is an exact match or relatively nearer match that pixel coordinate is stored.

finally, mean of the stored pixel coordinates may result in a straight line as i have shown.

so far i have been using this code to do this.

load clown
M = mean(X);
figure, imshow(X,[]); 
hold on;
down=zeros(1,numel(M));

 for columnIdx = 1:numel(M)
    meanValue = M(columnIdx);

    % find locations of gray-scale values equal to the mean        
    idx = find(X(:, columnIdx) == meanValue);

    % if not exact match find relative near alues ~ to the mean
    if (isempty(idx) == 1)
        [~, idx] = min(dist(X(:,columnIdx),meanValue));
    end

    % select one value & make a vector for plotting
    down(1,columnIdx)=idx(1);
end
% plot
plot([1, numel(M)], [mean(down), mean(down)], 'r-')
hold off;

it will result in the following output,

actually what i am interested in are the 'y' pixel coordinates so that i can divide the image into 2 parts. superimposing plots cannot do this i guess. please correct me if my intuition is wrong or if possible can you please suggest a simpler way to do this. thanks :)

[Guillaume]

Well, you seem to have answered your own question there. Doesn't the algorithm you wrote do what you want?

The only thing I would change is the inside of your for loop to:

 meanValue = M(columnIdx);
 [~, idx] = min(abs(X(:, columnIdx) - meanValue);
 down(1, columnIdx) = idx;

[Yawar Rehman]

i am not sure it is the correct answer though it is doing what i want! + i am curious if someone could suggest me a way for avoiding for loop

Answer 1

[Guillaume]

To do what you want without a loop:

 [~, down] = min(abs(bsxfun(@minus, X, mean(X))));

bsxfun just subtract the column mean from every value in X, and then you take the index of the minimum of the absolute value of that difference.

[Yawar Rehman]

kudos ... thank you guillaume

Answer 2

[Michael Haderlein]

Actually, I believe he just gave exactly contradicting answers on Guillaume's question. As I get it, this is what the result should look like:

I did it with the following code, however, I have no idea why a more straight-forward way didn't work (maybe someone can tell me?):

img=imread('grayscaleexample.bmp');
figure, imshow(img)
ha=axes('color','none','position',get(gca,'position'));
tf=figure; 
hp=plot(1:size(img,2),mean(img)); 
set(hp,'parent',ha)
set(ha,'xlim',[1,size(img,2)])
delete(tf)

So the idea is to create another axes above the ones showing the image and plot the mean of the grayscale there. For some reason (please help) I have to draw the plot somewhere else first and then copy it to the axes where it's shown finally. When I plot directly in the axes on top of the ones with the image, the image axes will be deleted - I have no idea why.

What I mean is the following:

>> figure, imshow(img)
>> axes('position',get(gca,'position'))
>> get(gcf,'children')

ans =

     1.0e+03 *

      1.4420
      1.4370

>> plot(1:size(img,2),mean(img));
>> get(gcf,'children')

ans =

     1.4370e+03

[Guillaume]

plot will always create a new axis unless you precede it with:

 hold on

or

 hold all

[Michael Haderlein]

ok, but why are the other axes deleted?

[Michael Haderlein]

Anyway, thanks for reminding me on hold. I wasn't aware that plot actually creates new axes but thought it would delete the children of the current axes. Thus, it goes much easier with:

 figure, imshow(img)
 axes('position',get(gca,'position'))
 hold all
 plot(1:size(img,2),mean(img));
 set(gca,'color','none','xlim',[1 size(img,2)])

[Guillaume]

It's not deleted. You just can't see the old one because of the white background of the new one. If you used:

 axes('position',get(gca,'position'), 'Color', 'none')

You'd see the old axis since the new one is transparent.

Anyway, you don't need to create a new axis.

[Guillaume]

All the high level plotting functions (plot, bar, surf, etc.) create a new axis.

[Michael Haderlein]

Yes, they create new axes if no axes are there and they also reset part of the axes properties in case there are some already. However, in this example, the old axes really were deleted as you see in my example. It has nothing to do with the background.

[Yawar Rehman]

thank for the answer haderlein. i have commented on the question which i asked. can you please review it

Answer 3

[Guillaume]

If you do indeed want to plot the mean on the graph (i.e. the other option I offered) then:

 imshow(img);
 hold on
 plot(1:size(img, 2), mean(img), 'r');

I'm not sure what you mean by having two portions of the image though.

[Michael Haderlein]

oops, you're totally right with that, of course plotting in the same axes is possible. This axes deleting behavior confused me so much that I didn't think about this anymore ;-)