Ringing arises because most good synthesis functions are oscillating. In fact, ringing is less of an artifact — in the sense that it would correspond to a deterioration of the data — than the consequence of the choice of a model: it may sometimes happen (but this is not the rule) that data are represented (e.g., magnified or zoomed) in such a way that the representation, although appearing to be plagued with ringing "artifacts," is nonetheless exact and allows the perfect recovery of the initial samples. Ringing can also be highlighted by translating by a noninteger amount a signal where there is a localized domain of constant samples bordered by sharp edges. After interpolation, translation, and resampling, the new samples do no more exhibit a constant value over the translated domain, but they tend to oscillate. This is known as the Gibbs effect; its perceptual counterpart is the Mach bands phenomena. Figure 1 shows an occurrence of ringing in the outlined area due to the horizontal translation of a high-contrast image by half a pixel.
Was this article helpful?