'Pends on what "center" and "best" mean... :)
Normally, I'd say just
but on trying that, one gets a somewhat unexpected result:
The problem is your sampling is so overloaded to the lower RH quadrant --
It's not just a few over the average elsewhere but those four bins are like 8-10X the number of observations as the upper end of the others so the means are grossly weighted in that direction.
Again depending upon what is desired, you could either then visually select a shift of the origin offset or, with some effort, compute a weighted mean from the histogram counts and bin centers to counteract the samping discrepancy and shift the mean to more nearly the geometric centroid.
Or, you could try simply using the bounding box means...as a thought just comes to mind--
That's probably closer to what you had in mind...