In the corrected example that I provided, I have modified three points to get it right:
1) exchanged the indices "centers" 1 and 2
2) transposed counts'
3) made the y axis  (axis xy)

This is why I think the example quite is strange.

comment #2:

> figure
> [counts, centers] = hist3 (data);
> imagesc (centers{1}, centers{2}, counts')
> axis xy
> xlabel('X')
> ylabel('Y')

maybe someone more familiar can confirm, but i believe image() files may use a different (1,1) location than regular arrays.

yes it should be probably 'especially'.

this is part of the statistics package, which now handles code and issue tracking at

https://github.com/gnu-octave/statistics

looking at

demo hist3:

X = [
    1    1
    1    1
    1   10
    1   10
    5    5
    5    5
    5    5
    5    5
    5    5
    7    3
    7    3
    7    3
   10   10
   10   10];

[N,C]=hist3(X)
N =

   2   0   0   0   0   0   0   0   0   2
   0   0   0   0   0   0   0   0   0   0
   0   0   0   0   0   0   0   0   0   0
   0   0   0   0   0   0   0   0   0   0
   0   0   0   0   5   0   0   0   0   0
   0   0   0   0   0   0   0   0   0   0
   0   0   3   0   0   0   0   0   0   0
   0   0   0   0   0   0   0   0   0   0
   0   0   0   0   0   0   0   0   0   0
   0   0   0   0   0   0   0   0   0   2

C =
{
  [1,1] =

     1.4500   2.3500   3.2500   4.1500   5.0500   5.9500   6.8500   7.7500   8.6500   9.5500

  [1,2] =

     1.4500   2.3500   3.2500   4.1500   5.0500   5.9500   6.8500   7.7500   8.6500   9.5500

C's look ok, but they're uniform so it's tough to tell if they're flipped.

looking at N, the three corners with data in the plot look like they match the three non-zero corner bins. is the issue that images use different array zeros?