By measuring whether the image is circularly symmetric.
Fair enough, but if you’re trying to measure beyond the diffracttion limit, it will be circularly symmetric, whether there are two sources or one.
OK, I was talking out of my hat. Ignorance fought. If for an ideal lens the MTF is the autocorrelation of the aperture, it is clear that it would go absolutely to zero at a finite spatial frequency.
However, I think the second half of my post is true. I was led astray by falsely generalizing it. For a single point source, you can localize it’s position with arbitrarily fine resolution if you have high enough signal to noise. This is the basis for the super resolution techniques known by acronyms like PALM and STORM.