Okay, there's some issue, though not exactly a direct bug in the algorithm. Here's the deal:

Histogram.newFromARGBArray() will internally run ColorList.createFromARGBArray() to provide a list of samples. This function however will always pick a given number of RANDOM samples of the image. This behaviour is documented, but every time you'll get slightly different histograms as a result, even though they will all converge to a very small difference/error as you increase the number of samples to the number of pixels in the source image.

To avoid the random sampling of the  createFromARGBArray() function, there's actually another constructor for the Histogram which will take an already existing ColorList as parameter. And the ColorList itself has a constructor to take a pixel array, so in theory you should be able to do the following:

1. Histogram hist=new Histogram(new ColorList(img.pixels));
2. hist.compute(tolerance,blendCols);

BUT... The issue is that I somehow have overlooked to declare the compute() method to be "public", so you can't currently call it from code outside the toxi.color package (like your sketch).

BUT #2: As a temporary workaround until the 0021 release is out, we can cheat and create a new tab in the Processing PDE and name it "MyHist.java" and pretend this class below is part of the toxi.color java package. So just copy this chunk in the new tab & save it.

1. package toxi.color;
2. 
   
3. import java.util.List;
4. 
   
5. public class MyHist extends Histogram {
6.   
7.   public MyHist(ColorList cols) {
8.     super(cols);
9.   }
10.   
11.   public List<HistEntry> compute(float tolerance, boolean blendCols) { 
12.     return super.compute(tolerance,blendCols);
13.   }
14. }

Lastly, you only need to update your sketch to use "MyHist" and not the usual "Histogram":

1. Histogram hist=new MyHist(new ColorList(img.pixels));
2. hist.compute(tolerance,true); // or false, for no blending of binned colors

Btw. This visibility issue is already fixed now and will require this hack in the near future when the next release is out

Thnx for your very fast reply.

I will try your solution very soon.

 

But an additional question for my understanding,

You say:

"even though they will all converge to a very small difference/error as you increase the number of samples to the number of pixels in the source image."

 

I am not sure if I understand that.

 

What I did was just

- Run the ImageColorExample 100 times

- changed

 Histogram hist=Histogram.newFromARGBArray(img.pixels, img.pixels.length/10, tolerance, true);

to (means to me I take all pixels)

 Histogram hist=Histogram.newFromARGBArray(img.pixels, img.pixels.length, tolerance, true);

- println(e.getColor().toHex()+": "+e.getFrequency()+": "+hist.getEntries().size());

 

And I can see differences in all 3 values in the order of 10%. Would you expect that  or is there still a misunderstanding

on my side ?

 

Ok, it works. Thnx again a lot for your help !

Little Typo in you code:

MyHist hist=new MyHist(new ColorList(img.pixels));