Sharpness and aliasing, one more time

The question of whether aliasing counts as “fake sharpness” comes up regularly in discussions of camera resolution. In a recent post I summarized the difference between the GFX 50x and GFX 100x this way: “images are just as sharp, one has just more aliasing.” Someone objected that this must be logically flawed because aliasing creates “fake detail,” which, in their view, cannot be as sharp as real captured detail. That position feels intuitive, but it collapses once you look at sharpness the way I do. The usual way to quantify digital camera system sharpness is the modulation transfer function, plotted in cycles per picture height. And MTF separates sharpness from aliasing rather than blending the two together.

A clean example comes from comparing Fujifilm’s GFX 50x and GFX 100x. Both sensors use nearly identical pixel apertures. If you put the same lens on both cameras, use the same aperture and distance, and look at MTF in cycles per picture height, the curves match closely from zero spatial frequency up to the Nyquist limit of the GFX 50x. In that region, both sensors are sampling the same optical field and capturing contrast with essentially the same efficiency. They are equally sharp over that band of frequencies.

The difference appears only once you move past the 50x Nyquist frequency. The GFX 100x, with its finer pitch, has a higher Nyquist limit. Spatial content that falls between the two Nyquist frequencies is sampled cleanly by the 100x and is aliased by the 50x. The 50x cannot recover that information because it cannot sample it adequately. The 100x can. But this gap in recoverable detail says nothing about how sharp the 50x is below its Nyquist threshold. It only describes what each system can and cannot reconstruct at higher spatial frequencies.

This is why aliasing and sharpness need to be treated as orthogonal ideas. Sharpness is measured as contrast transfer at a particular frequency. Aliasing is what happens when the system tries to reconstruct frequencies it cannot sample. Below the 50x Nyquist boundary, both cameras are equally sharp because their MTFs are essentially identical. Above that boundary, only the 100x can produce accurate information. The 50x produces structured artifacts. Those artifacts have nothing to do with sharpness. They are not degraded sharpness, and they are not enhanced sharpness. They are simply the appearance of undersampling.

It is therefore misleading to say that an image “cannot be as sharp” if it contains aliasing. It can be just as sharp over the frequencies the system can sample, and the presence or absence of aliasing at higher frequencies does not retroactively alter that. Above Nyquist a camera may still produce crisp-looking structure, but that structure is not correct scene information, because the sampling theorem has been violated. Aliasing does not reduce sharpness, and it does not increase it. It occupies a different conceptual axis.

The practical outcome is straightforward. Two systems with matched MTFs up to a given frequency will look the same in that range. A system with a higher Nyquist limit will continue to deliver real detail beyond the point where the lower-resolution system substitutes artifacts. Those artifacts do not make the lower-resolution system less sharp where it is operating within its Nyquist bandwidth. They only reveal the point where that bandwidth runs out.

This distinction becomes important when comparing cameras like the GFX 50x and GFX 100x or when deciding how large a print a given file can support. Aliasing is not fake sharpness. It is fake content produced when the sampling capacity of the system has been exceeded.