aliasing

Aliasing is what goes wrong when you sample a scene too coarsely to capture its fine detail: the detail does not disappear, it disguises itself as a coarse, false pattern that was never in the scene. The classic everyday version is the wagon-wheel effect, where a fast-spinning wheel filmed at a slower frame rate appears to rotate slowly or backward. In still images it shows up as moire (shimmering ripples across fine fabric or distant brick) and as jagged staircase edges (jaggies) along smooth diagonal lines.

The cause is undersampling relative to the Nyquist limit. The sampling theorem says you must take more than two samples per cycle of the highest spatial frequency present; when a scene contains frequencies above half the sampling rate, those frequencies fold back and are indistinguishable from genuine lower frequencies — different inputs become aliases of one another, hence the name. Because the false pattern is now a low frequency, it cannot be removed after sampling: once the data is recorded, the high-frequency original and its low-frequency alias are mathematically the same samples, so no amount of post-processing can separate them.

Since you cannot fix it afterward, aliasing must be prevented before sampling by anti-aliasing: low-pass filter (blur) the signal to remove frequencies above the Nyquist limit, then sample. Cameras use an optical anti-aliasing filter in front of the sensor and integrate light over each pixel's finite area; renderers supersample (render at higher resolution and average down) or use multisample anti-aliasing; correct image downscaling blurs before subsampling. The same principle pervades deep vision: strided convolutions and pooling are downsampling steps that can alias, which is why anti-aliased (blur-pooling) downsampling improves the shift-consistency of CNNs. Aliasing is therefore not a cosmetic nuisance but a fundamental signal-processing limit on what sampling can faithfully record.