barchart not appearing as it should

Why does my bar chart appear black and not blue as instructed to?

 [maxval,idx]=max(IM(:));
 maxval= double(maxval);
 [counts,x] = imhist(IM,maxval);
 bar(x,counts,'b');

thanks Jason

[Andrew Reibold]

It is blue for me using 2014a.

Try another color to make sure its not just that blue is similar to black on your monitor or something.

Answer 1

[Image Analyst]

Jason, It shows up as blue for me in R2014b. But if your IM is uint8 or uint16, your call to imhist is wrong - it will be aliased and you will have a false spike in the middle. Just run this to see it:

IM = uint8(randi(255, 256,256));
[maxval,idx]=max(IM(:));
maxval= double(maxval);
[counts,x] = imhist(IM,maxval);
bar(x,counts,'b');

To fix, do this:

IM = uint8(randi(255, 256,256));
[maxval,idx]=max(IM(:));
maxval= double(maxval);
[counts,x] = imhist(IM,maxval+1);
bar(x,counts,'b');

The reason is you need a bin for intensity 0. If you don't, and you have let's say values of 0-255 and you're using 255 bins, one of the bins has to be doubled up because you're using only 255 bins instead of 256 bins.

If you have a really old version of MATLAB, you can try the 'Color' option, which a lot of MATLAB functions still use:

bar(x,counts, 'Color', 'b');

Again, not necessary in R2014b, and may not even work for older releases either, but it's easy enough to try.

[Jason]

Thanks IM. Its still not displaying correctly - wrong colour and strange "periodic" peaks. the stem plot displays the correct colour.

Two things, my image cis 16bit and the max val could be 65535. Does the imhist allow greater than 16bit values i.e. 65536?

Also, does this mean then if I am wanting to plot the mode on this histogram, i have to do this:

xm=mode(IM(:))

plot([xm xm],[0 max(counts)],'r', 'LineWidth',1);

or is the mode really xm+1 ??

I don't like the spikes, it is some sort of aliasing? thanks

[Image Analyst]

Yes. Exactly. The spikes are aliasing. The problem is that you have one less bin than the number of gray levels and so to squeeze them all into the fewer number of bins that you are using, some bins will have to double up. For example if your highest gray level was 30,000 and that means that you want gray level bins from 0 to 30,000. Well, that would be 30,001 bins, not 30,000 bins. So if you force it to use 30,000 bins instead of 30,001, then some bin is going to collect counts from two bins and that's what shows up as spikes. That's what I tried to explain in my answer. Actually if you just don't even pass in the number of bins, it will figure it out perfectly. It will use 65536 bins. Then to avoid a bunch of zero bins on the right, you can just adjust the scale with xlim():

[counts, grayLevels] = imhist(array16Bit);
bar(grayLevels, counts);
xlim([0, maxGrayLevel]);

I hope that explains it. If so, thanks in advance for officially "Accepting" the answer.

[Jason]

Im going to just hardwire the number of bins: Its still not colouring blue (I have 2014b);

[counts,x] = imhist(IM,65536);   
 grid on;
 hold on;
 bar(x,counts,'b');

[Jason]

Got it!!!

When you have so many bars, you also need to change the edge colour:

   bar(x,counts,'b','EdgeColor','b');

[Image Analyst]

Glad we got the aliasing and color problems solved. If the histogram is merely for display there's no sense in making more bins than the number of pixels across your screen so you might want to reduce that. Of course if you need it for some kind of autothresholding algorithm, then you can keep all of them. Maybe you should go ahead and mark this question as "Answered" now, if I've helped you.

[Jason]

Hi, I was going to start a new thread regarding reducing the bins. Would you mind if I ask here.

I agree, I only need say 1000 for display purposes. But I want the xlim max to be the intensity that 95% of all pixels have or are below.

I thought this should work but its not quite right:

    nbins=65536;
    [counts,x] = imhist(IM,nbins);
    sumprod=0;
    for i=1:numel(counts);
        sumcurrent(i)=counts(i)*x(i);
        sumprod=sumprod+sumcurrent(i);
        if sumprod > target
          break
     end

end

Im sure I need to include nbins somewhere for when each bin isn't unity intensity value??

[Image Analyst]

Just use cumsum() and find():

cdf = cumsum(counts); % Cumulative Distribution Function.
cdf = cdf / cdf(end); % Normalize to 0-1.
% Find index of 95%
index95 = find(cdf > 0.95, 1, 'first');
% Find gray level of 95%
grayLevel95 = x(index95);

[Jason]

Thankyou